v8/src/full-codegen.cc

1589 lines
49 KiB
C++
Raw Normal View History

// Copyright 2012 the V8 project authors. All rights reserved.
Initial infrastructure for fast compilation of top-level code. The fast code generator is optimized for compilation time and code size. Currently it is only implemented on IA32. It is potentially triggered for any code in the global scope (including code eval'd in the global scope). It performs a syntactic check and chooses to compile in fast mode if the AST contains only supported constructs and matches some other constraints. Initially supported constructs are * ExpressionStatement, * ReturnStatement, * VariableProxy (variable references) to parameters and stack-allocated locals, * Assignment with lhs a parameter or stack-allocated local, and * Literal This allows compilation of literals at the top level and not much else. All intermediate values are allocated to temporaries and the stack is used for all temporaries. The extra memory traffic is a known issue. The code generated for 'true' is: 0 push ebp 1 mov ebp,esp 3 push esi 4 push edi 5 push 0xf5cca135 ;; object: 0xf5cca135 <undefined> 10 cmp esp,[0x8277efc] 16 jnc 27 (0xf5cbbb1b) 22 call 0xf5cac960 ;; code: STUB, StackCheck, minor: 0 27 push 0xf5cca161 ;; object: 0xf5cca161 <true> 32 mov eax,[esp] 35 mov [ebp+0xf4],eax 38 pop eax 39 mov eax,[ebp+0xf4] 42 mov esp,ebp ;; js return 44 pop ebp 45 ret 0x4 48 mov eax,0xf5cca135 ;; object: 0xf5cca135 <undefined> 53 mov esp,ebp ;; js return 55 pop ebp 56 ret 0x4 Review URL: http://codereview.chromium.org/273050 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@3067 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-10-14 19:30:50 +00:00
// Redistribution and use in source and binary forms, with or without
// modification, are permitted provided that the following conditions are
// met:
//
// * Redistributions of source code must retain the above copyright
// notice, this list of conditions and the following disclaimer.
// * Redistributions in binary form must reproduce the above
// copyright notice, this list of conditions and the following
// disclaimer in the documentation and/or other materials provided
// with the distribution.
// * Neither the name of Google Inc. nor the names of its
// contributors may be used to endorse or promote products derived
// from this software without specific prior written permission.
//
// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include "v8.h"
#include "codegen.h"
#include "compiler.h"
#include "debug.h"
#include "full-codegen.h"
#include "liveedit.h"
#include "macro-assembler.h"
#include "prettyprinter.h"
#include "scopes.h"
#include "scopeinfo.h"
#include "snapshot.h"
#include "stub-cache.h"
Initial infrastructure for fast compilation of top-level code. The fast code generator is optimized for compilation time and code size. Currently it is only implemented on IA32. It is potentially triggered for any code in the global scope (including code eval'd in the global scope). It performs a syntactic check and chooses to compile in fast mode if the AST contains only supported constructs and matches some other constraints. Initially supported constructs are * ExpressionStatement, * ReturnStatement, * VariableProxy (variable references) to parameters and stack-allocated locals, * Assignment with lhs a parameter or stack-allocated local, and * Literal This allows compilation of literals at the top level and not much else. All intermediate values are allocated to temporaries and the stack is used for all temporaries. The extra memory traffic is a known issue. The code generated for 'true' is: 0 push ebp 1 mov ebp,esp 3 push esi 4 push edi 5 push 0xf5cca135 ;; object: 0xf5cca135 <undefined> 10 cmp esp,[0x8277efc] 16 jnc 27 (0xf5cbbb1b) 22 call 0xf5cac960 ;; code: STUB, StackCheck, minor: 0 27 push 0xf5cca161 ;; object: 0xf5cca161 <true> 32 mov eax,[esp] 35 mov [ebp+0xf4],eax 38 pop eax 39 mov eax,[ebp+0xf4] 42 mov esp,ebp ;; js return 44 pop ebp 45 ret 0x4 48 mov eax,0xf5cca135 ;; object: 0xf5cca135 <undefined> 53 mov esp,ebp ;; js return 55 pop ebp 56 ret 0x4 Review URL: http://codereview.chromium.org/273050 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@3067 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-10-14 19:30:50 +00:00
namespace v8 {
namespace internal {
void BreakableStatementChecker::Check(Statement* stmt) {
Visit(stmt);
}
void BreakableStatementChecker::Check(Expression* expr) {
Visit(expr);
}
void BreakableStatementChecker::VisitVariableDeclaration(
VariableDeclaration* decl) {
}
void BreakableStatementChecker::VisitFunctionDeclaration(
FunctionDeclaration* decl) {
}
void BreakableStatementChecker::VisitModuleDeclaration(
ModuleDeclaration* decl) {
}
void BreakableStatementChecker::VisitImportDeclaration(
ImportDeclaration* decl) {
}
void BreakableStatementChecker::VisitExportDeclaration(
ExportDeclaration* decl) {
}
void BreakableStatementChecker::VisitModuleLiteral(ModuleLiteral* module) {
}
void BreakableStatementChecker::VisitModuleVariable(ModuleVariable* module) {
}
void BreakableStatementChecker::VisitModulePath(ModulePath* module) {
}
void BreakableStatementChecker::VisitModuleUrl(ModuleUrl* module) {
}
Get rid of static module allocation, do it in code. Modules now have their own local scope, represented by their own context. Module instance objects have an accessor for every export that forwards access to the respective slot from the module's context. (Exports that are modules themselves, however, are simple data properties.) All modules have a _hosting_ scope/context, which (currently) is the (innermost) enclosing global scope. To deal with recursion, nested modules are hosted by the same scope as global ones. For every (global or nested) module literal, the hosting context has an internal slot that points directly to the respective module context. This enables quick access to (statically resolved) module members by 2-dimensional access through the hosting context. For example, module A { let x; module B { let y; } } module C { let z; } allocates contexts as follows: [header| .A | .B | .C | A | C ] (global) | | | | | +-- [header| z ] (module) | | | +------- [header| y ] (module) | +------------ [header| x | B ] (module) Here, .A, .B, .C are the internal slots pointing to the hosted module contexts, whereas A, B, C hold the actual instance objects (note that every module context also points to the respective instance object through its extension slot in the header). To deal with arbitrary recursion and aliases between modules, they are created and initialized in several stages. Each stage applies to all modules in the hosting global scope, including nested ones. 1. Allocate: for each module _literal_, allocate the module contexts and respective instance object and wire them up. This happens in the PushModuleContext runtime function, as generated by AllocateModules (invoked by VisitDeclarations in the hosting scope). 2. Bind: for each module _declaration_ (i.e. literals as well as aliases), assign the respective instance object to respective local variables. This happens in VisitModuleDeclaration, and uses the instance objects created in the previous stage. For each module _literal_, this phase also constructs a module descriptor for the next stage. This happens in VisitModuleLiteral. 3. Populate: invoke the DeclareModules runtime function to populate each _instance_ object with accessors for it exports. This is generated by DeclareModules (invoked by VisitDeclarations in the hosting scope again), and uses the descriptors generated in the previous stage. 4. Initialize: execute the module bodies (and other code) in sequence. This happens by the separate statements generated for module bodies. To reenter the module scopes properly, the parser inserted ModuleStatements. R=mstarzinger@chromium.org,svenpanne@chromium.org BUG= Review URL: https://codereview.chromium.org/11093074 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@13033 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2012-11-22 10:25:22 +00:00
void BreakableStatementChecker::VisitModuleStatement(ModuleStatement* stmt) {
}
void BreakableStatementChecker::VisitBlock(Block* stmt) {
}
void BreakableStatementChecker::VisitExpressionStatement(
ExpressionStatement* stmt) {
// Check if expression is breakable.
Visit(stmt->expression());
}
void BreakableStatementChecker::VisitEmptyStatement(EmptyStatement* stmt) {
}
void BreakableStatementChecker::VisitIfStatement(IfStatement* stmt) {
// If the condition is breakable the if statement is breakable.
Visit(stmt->condition());
}
void BreakableStatementChecker::VisitContinueStatement(
ContinueStatement* stmt) {
}
void BreakableStatementChecker::VisitBreakStatement(BreakStatement* stmt) {
}
void BreakableStatementChecker::VisitReturnStatement(ReturnStatement* stmt) {
// Return is breakable if the expression is.
Visit(stmt->expression());
}
void BreakableStatementChecker::VisitWithStatement(WithStatement* stmt) {
Visit(stmt->expression());
}
void BreakableStatementChecker::VisitSwitchStatement(SwitchStatement* stmt) {
// Switch statements breakable if the tag expression is.
Visit(stmt->tag());
}
void BreakableStatementChecker::VisitDoWhileStatement(DoWhileStatement* stmt) {
// Mark do while as breakable to avoid adding a break slot in front of it.
is_breakable_ = true;
}
void BreakableStatementChecker::VisitWhileStatement(WhileStatement* stmt) {
// Mark while statements breakable if the condition expression is.
Visit(stmt->cond());
}
void BreakableStatementChecker::VisitForStatement(ForStatement* stmt) {
// Mark for statements breakable if the condition expression is.
if (stmt->cond() != NULL) {
Visit(stmt->cond());
}
}
void BreakableStatementChecker::VisitForInStatement(ForInStatement* stmt) {
// Mark for in statements breakable if the enumerable expression is.
Visit(stmt->enumerable());
}
void BreakableStatementChecker::VisitTryCatchStatement(
TryCatchStatement* stmt) {
// Mark try catch as breakable to avoid adding a break slot in front of it.
is_breakable_ = true;
}
void BreakableStatementChecker::VisitTryFinallyStatement(
TryFinallyStatement* stmt) {
// Mark try finally as breakable to avoid adding a break slot in front of it.
is_breakable_ = true;
}
void BreakableStatementChecker::VisitDebuggerStatement(
DebuggerStatement* stmt) {
// The debugger statement is breakable.
is_breakable_ = true;
}
void BreakableStatementChecker::VisitFunctionLiteral(FunctionLiteral* expr) {
}
void BreakableStatementChecker::VisitSharedFunctionInfoLiteral(
SharedFunctionInfoLiteral* expr) {
}
void BreakableStatementChecker::VisitConditional(Conditional* expr) {
}
void BreakableStatementChecker::VisitVariableProxy(VariableProxy* expr) {
}
void BreakableStatementChecker::VisitLiteral(Literal* expr) {
}
void BreakableStatementChecker::VisitRegExpLiteral(RegExpLiteral* expr) {
}
void BreakableStatementChecker::VisitObjectLiteral(ObjectLiteral* expr) {
}
void BreakableStatementChecker::VisitArrayLiteral(ArrayLiteral* expr) {
}
void BreakableStatementChecker::VisitAssignment(Assignment* expr) {
// If assigning to a property (including a global property) the assignment is
// breakable.
VariableProxy* proxy = expr->target()->AsVariableProxy();
Property* prop = expr->target()->AsProperty();
if (prop != NULL || (proxy != NULL && proxy->var()->IsUnallocated())) {
is_breakable_ = true;
return;
}
// Otherwise the assignment is breakable if the assigned value is.
Visit(expr->value());
}
void BreakableStatementChecker::VisitThrow(Throw* expr) {
// Throw is breakable if the expression is.
Visit(expr->exception());
}
void BreakableStatementChecker::VisitProperty(Property* expr) {
// Property load is breakable.
is_breakable_ = true;
}
void BreakableStatementChecker::VisitCall(Call* expr) {
// Function calls both through IC and call stub are breakable.
is_breakable_ = true;
}
void BreakableStatementChecker::VisitCallNew(CallNew* expr) {
// Function calls through new are breakable.
is_breakable_ = true;
}
void BreakableStatementChecker::VisitCallRuntime(CallRuntime* expr) {
}
void BreakableStatementChecker::VisitUnaryOperation(UnaryOperation* expr) {
Visit(expr->expression());
}
void BreakableStatementChecker::VisitCountOperation(CountOperation* expr) {
Visit(expr->expression());
}
void BreakableStatementChecker::VisitBinaryOperation(BinaryOperation* expr) {
Visit(expr->left());
if (expr->op() != Token::AND &&
expr->op() != Token::OR) {
Visit(expr->right());
}
}
void BreakableStatementChecker::VisitCompareOperation(CompareOperation* expr) {
Visit(expr->left());
Visit(expr->right());
}
void BreakableStatementChecker::VisitThisFunction(ThisFunction* expr) {
}
#define __ ACCESS_MASM(masm())
bool FullCodeGenerator::MakeCode(CompilationInfo* info) {
Isolate* isolate = info->isolate();
Handle<Script> script = info->script();
if (!script->IsUndefined() && !script->source()->IsUndefined()) {
int len = String::cast(script->source())->length();
isolate->counters()->total_full_codegen_source_size()->Increment(len);
}
if (FLAG_trace_codegen) {
PrintF("Full Compiler - ");
}
CodeGenerator::MakeCodePrologue(info);
Initial infrastructure for fast compilation of top-level code. The fast code generator is optimized for compilation time and code size. Currently it is only implemented on IA32. It is potentially triggered for any code in the global scope (including code eval'd in the global scope). It performs a syntactic check and chooses to compile in fast mode if the AST contains only supported constructs and matches some other constraints. Initially supported constructs are * ExpressionStatement, * ReturnStatement, * VariableProxy (variable references) to parameters and stack-allocated locals, * Assignment with lhs a parameter or stack-allocated local, and * Literal This allows compilation of literals at the top level and not much else. All intermediate values are allocated to temporaries and the stack is used for all temporaries. The extra memory traffic is a known issue. The code generated for 'true' is: 0 push ebp 1 mov ebp,esp 3 push esi 4 push edi 5 push 0xf5cca135 ;; object: 0xf5cca135 <undefined> 10 cmp esp,[0x8277efc] 16 jnc 27 (0xf5cbbb1b) 22 call 0xf5cac960 ;; code: STUB, StackCheck, minor: 0 27 push 0xf5cca161 ;; object: 0xf5cca161 <true> 32 mov eax,[esp] 35 mov [ebp+0xf4],eax 38 pop eax 39 mov eax,[ebp+0xf4] 42 mov esp,ebp ;; js return 44 pop ebp 45 ret 0x4 48 mov eax,0xf5cca135 ;; object: 0xf5cca135 <undefined> 53 mov esp,ebp ;; js return 55 pop ebp 56 ret 0x4 Review URL: http://codereview.chromium.org/273050 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@3067 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-10-14 19:30:50 +00:00
const int kInitialBufferSize = 4 * KB;
MacroAssembler masm(info->isolate(), NULL, kInitialBufferSize);
#ifdef ENABLE_GDB_JIT_INTERFACE
masm.positions_recorder()->StartGDBJITLineInfoRecording();
#endif
FullCodeGenerator cgen(&masm, info);
cgen.Generate();
Initial infrastructure for fast compilation of top-level code. The fast code generator is optimized for compilation time and code size. Currently it is only implemented on IA32. It is potentially triggered for any code in the global scope (including code eval'd in the global scope). It performs a syntactic check and chooses to compile in fast mode if the AST contains only supported constructs and matches some other constraints. Initially supported constructs are * ExpressionStatement, * ReturnStatement, * VariableProxy (variable references) to parameters and stack-allocated locals, * Assignment with lhs a parameter or stack-allocated local, and * Literal This allows compilation of literals at the top level and not much else. All intermediate values are allocated to temporaries and the stack is used for all temporaries. The extra memory traffic is a known issue. The code generated for 'true' is: 0 push ebp 1 mov ebp,esp 3 push esi 4 push edi 5 push 0xf5cca135 ;; object: 0xf5cca135 <undefined> 10 cmp esp,[0x8277efc] 16 jnc 27 (0xf5cbbb1b) 22 call 0xf5cac960 ;; code: STUB, StackCheck, minor: 0 27 push 0xf5cca161 ;; object: 0xf5cca161 <true> 32 mov eax,[esp] 35 mov [ebp+0xf4],eax 38 pop eax 39 mov eax,[ebp+0xf4] 42 mov esp,ebp ;; js return 44 pop ebp 45 ret 0x4 48 mov eax,0xf5cca135 ;; object: 0xf5cca135 <undefined> 53 mov esp,ebp ;; js return 55 pop ebp 56 ret 0x4 Review URL: http://codereview.chromium.org/273050 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@3067 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-10-14 19:30:50 +00:00
if (cgen.HasStackOverflow()) {
ASSERT(!isolate->has_pending_exception());
return false;
Initial infrastructure for fast compilation of top-level code. The fast code generator is optimized for compilation time and code size. Currently it is only implemented on IA32. It is potentially triggered for any code in the global scope (including code eval'd in the global scope). It performs a syntactic check and chooses to compile in fast mode if the AST contains only supported constructs and matches some other constraints. Initially supported constructs are * ExpressionStatement, * ReturnStatement, * VariableProxy (variable references) to parameters and stack-allocated locals, * Assignment with lhs a parameter or stack-allocated local, and * Literal This allows compilation of literals at the top level and not much else. All intermediate values are allocated to temporaries and the stack is used for all temporaries. The extra memory traffic is a known issue. The code generated for 'true' is: 0 push ebp 1 mov ebp,esp 3 push esi 4 push edi 5 push 0xf5cca135 ;; object: 0xf5cca135 <undefined> 10 cmp esp,[0x8277efc] 16 jnc 27 (0xf5cbbb1b) 22 call 0xf5cac960 ;; code: STUB, StackCheck, minor: 0 27 push 0xf5cca161 ;; object: 0xf5cca161 <true> 32 mov eax,[esp] 35 mov [ebp+0xf4],eax 38 pop eax 39 mov eax,[ebp+0xf4] 42 mov esp,ebp ;; js return 44 pop ebp 45 ret 0x4 48 mov eax,0xf5cca135 ;; object: 0xf5cca135 <undefined> 53 mov esp,ebp ;; js return 55 pop ebp 56 ret 0x4 Review URL: http://codereview.chromium.org/273050 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@3067 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-10-14 19:30:50 +00:00
}
unsigned table_offset = cgen.EmitStackCheckTable();
Code::Flags flags = Code::ComputeFlags(Code::FUNCTION);
Handle<Code> code = CodeGenerator::MakeCodeEpilogue(&masm, flags, info);
code->set_optimizable(info->IsOptimizable() &&
!info->function()->flags()->Contains(kDontOptimize) &&
info->function()->scope()->AllowsLazyCompilation());
cgen.PopulateDeoptimizationData(code);
cgen.PopulateTypeFeedbackInfo(code);
cgen.PopulateTypeFeedbackCells(code);
code->set_has_deoptimization_support(info->HasDeoptimizationSupport());
code->set_handler_table(*cgen.handler_table());
#ifdef ENABLE_DEBUGGER_SUPPORT
code->set_has_debug_break_slots(
info->isolate()->debugger()->IsDebuggerActive());
code->set_compiled_optimizable(info->IsOptimizable());
#endif // ENABLE_DEBUGGER_SUPPORT
code->set_allow_osr_at_loop_nesting_level(0);
code->set_profiler_ticks(0);
code->set_stack_check_table_offset(table_offset);
CodeGenerator::PrintCode(code, info);
info->SetCode(code); // May be an empty handle.
#ifdef ENABLE_GDB_JIT_INTERFACE
if (FLAG_gdbjit && !code.is_null()) {
GDBJITLineInfo* lineinfo =
masm.positions_recorder()->DetachGDBJITLineInfo();
GDBJIT(RegisterDetailedLineInfo(*code, lineinfo));
}
#endif
return !code.is_null();
Initial infrastructure for fast compilation of top-level code. The fast code generator is optimized for compilation time and code size. Currently it is only implemented on IA32. It is potentially triggered for any code in the global scope (including code eval'd in the global scope). It performs a syntactic check and chooses to compile in fast mode if the AST contains only supported constructs and matches some other constraints. Initially supported constructs are * ExpressionStatement, * ReturnStatement, * VariableProxy (variable references) to parameters and stack-allocated locals, * Assignment with lhs a parameter or stack-allocated local, and * Literal This allows compilation of literals at the top level and not much else. All intermediate values are allocated to temporaries and the stack is used for all temporaries. The extra memory traffic is a known issue. The code generated for 'true' is: 0 push ebp 1 mov ebp,esp 3 push esi 4 push edi 5 push 0xf5cca135 ;; object: 0xf5cca135 <undefined> 10 cmp esp,[0x8277efc] 16 jnc 27 (0xf5cbbb1b) 22 call 0xf5cac960 ;; code: STUB, StackCheck, minor: 0 27 push 0xf5cca161 ;; object: 0xf5cca161 <true> 32 mov eax,[esp] 35 mov [ebp+0xf4],eax 38 pop eax 39 mov eax,[ebp+0xf4] 42 mov esp,ebp ;; js return 44 pop ebp 45 ret 0x4 48 mov eax,0xf5cca135 ;; object: 0xf5cca135 <undefined> 53 mov esp,ebp ;; js return 55 pop ebp 56 ret 0x4 Review URL: http://codereview.chromium.org/273050 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@3067 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-10-14 19:30:50 +00:00
}
unsigned FullCodeGenerator::EmitStackCheckTable() {
// The stack check table consists of a length (in number of entries)
// field, and then a sequence of entries. Each entry is a pair of AST id
// and code-relative pc offset.
masm()->Align(kIntSize);
unsigned offset = masm()->pc_offset();
unsigned length = stack_checks_.length();
__ dd(length);
for (unsigned i = 0; i < length; ++i) {
__ dd(stack_checks_[i].id.ToInt());
__ dd(stack_checks_[i].pc_and_state);
}
return offset;
}
void FullCodeGenerator::PopulateDeoptimizationData(Handle<Code> code) {
// Fill in the deoptimization information.
ASSERT(info_->HasDeoptimizationSupport() || bailout_entries_.is_empty());
if (!info_->HasDeoptimizationSupport()) return;
int length = bailout_entries_.length();
Handle<DeoptimizationOutputData> data = isolate()->factory()->
NewDeoptimizationOutputData(length, TENURED);
for (int i = 0; i < length; i++) {
data->SetAstId(i, bailout_entries_[i].id);
data->SetPcAndState(i, Smi::FromInt(bailout_entries_[i].pc_and_state));
}
code->set_deoptimization_data(*data);
}
void FullCodeGenerator::PopulateTypeFeedbackInfo(Handle<Code> code) {
Handle<TypeFeedbackInfo> info = isolate()->factory()->NewTypeFeedbackInfo();
info->set_ic_total_count(ic_total_count_);
ASSERT(!isolate()->heap()->InNewSpace(*info));
code->set_type_feedback_info(*info);
}
void FullCodeGenerator::Initialize() {
// The generation of debug code must match between the snapshot code and the
// code that is generated later. This is assumed by the debugger when it is
// calculating PC offsets after generating a debug version of code. Therefore
// we disable the production of debug code in the full compiler if we are
// either generating a snapshot or we booted from a snapshot.
generate_debug_code_ = FLAG_debug_code &&
!Serializer::enabled() &&
!Snapshot::HaveASnapshotToStartFrom();
masm_->set_emit_debug_code(generate_debug_code_);
masm_->set_predictable_code_size(true);
InitializeAstVisitor();
}
void FullCodeGenerator::PopulateTypeFeedbackCells(Handle<Code> code) {
if (type_feedback_cells_.is_empty()) return;
int length = type_feedback_cells_.length();
int array_size = TypeFeedbackCells::LengthOfFixedArray(length);
Handle<TypeFeedbackCells> cache = Handle<TypeFeedbackCells>::cast(
isolate()->factory()->NewFixedArray(array_size, TENURED));
for (int i = 0; i < length; i++) {
cache->SetAstId(i, type_feedback_cells_[i].ast_id);
cache->SetCell(i, *type_feedback_cells_[i].cell);
}
TypeFeedbackInfo::cast(code->type_feedback_info())->set_type_feedback_cells(
*cache);
}
void FullCodeGenerator::PrepareForBailout(Expression* node, State state) {
PrepareForBailoutForId(node->id(), state);
}
void FullCodeGenerator::RecordJSReturnSite(Call* call) {
// We record the offset of the function return so we can rebuild the frame
// if the function was inlined, i.e., this is the return address in the
// inlined function's frame.
//
// The state is ignored. We defensively set it to TOS_REG, which is the
// real state of the unoptimized code at the return site.
PrepareForBailoutForId(call->ReturnId(), TOS_REG);
#ifdef DEBUG
// In debug builds, mark the return so we can verify that this function
// was called.
ASSERT(!call->return_is_recorded_);
call->return_is_recorded_ = true;
#endif
}
void FullCodeGenerator::PrepareForBailoutForId(BailoutId id, State state) {
// There's no need to prepare this code for bailouts from already optimized
// code or code that can't be optimized.
if (!info_->HasDeoptimizationSupport()) return;
unsigned pc_and_state =
StateField::encode(state) | PcField::encode(masm_->pc_offset());
ASSERT(Smi::IsValid(pc_and_state));
BailoutEntry entry = { id, pc_and_state };
#ifdef DEBUG
if (FLAG_enable_slow_asserts) {
// Assert that we don't have multiple bailout entries for the same node.
for (int i = 0; i < bailout_entries_.length(); i++) {
if (bailout_entries_.at(i).id == entry.id) {
AstPrinter printer;
PrintF("%s", printer.PrintProgram(info_->function()));
UNREACHABLE();
}
}
}
#endif // DEBUG
bailout_entries_.Add(entry, zone());
}
void FullCodeGenerator::RecordTypeFeedbackCell(
TypeFeedbackId id, Handle<JSGlobalPropertyCell> cell) {
TypeFeedbackCellEntry entry = { id, cell };
type_feedback_cells_.Add(entry, zone());
}
void FullCodeGenerator::RecordStackCheck(BailoutId ast_id) {
// The pc offset does not need to be encoded and packed together with a
// state.
ASSERT(masm_->pc_offset() > 0);
BailoutEntry entry = { ast_id, static_cast<unsigned>(masm_->pc_offset()) };
stack_checks_.Add(entry, zone());
}
bool FullCodeGenerator::ShouldInlineSmiCase(Token::Value op) {
// Inline smi case inside loops, but not division and modulo which
// are too complicated and take up too much space.
if (op == Token::DIV ||op == Token::MOD) return false;
if (FLAG_always_inline_smi_code) return true;
return loop_depth_ > 0;
}
void FullCodeGenerator::EffectContext::Plug(Register reg) const {
}
void FullCodeGenerator::AccumulatorValueContext::Plug(Register reg) const {
__ Move(result_register(), reg);
}
void FullCodeGenerator::StackValueContext::Plug(Register reg) const {
__ push(reg);
}
void FullCodeGenerator::TestContext::Plug(Register reg) const {
// For simplicity we always test the accumulator register.
__ Move(result_register(), reg);
codegen()->PrepareForBailoutBeforeSplit(condition(), false, NULL, NULL);
codegen()->DoTest(this);
}
void FullCodeGenerator::EffectContext::PlugTOS() const {
__ Drop(1);
}
void FullCodeGenerator::AccumulatorValueContext::PlugTOS() const {
__ pop(result_register());
}
void FullCodeGenerator::StackValueContext::PlugTOS() const {
}
void FullCodeGenerator::TestContext::PlugTOS() const {
// For simplicity we always test the accumulator register.
__ pop(result_register());
codegen()->PrepareForBailoutBeforeSplit(condition(), false, NULL, NULL);
codegen()->DoTest(this);
}
void FullCodeGenerator::EffectContext::PrepareTest(
Label* materialize_true,
Label* materialize_false,
Label** if_true,
Label** if_false,
Label** fall_through) const {
// In an effect context, the true and the false case branch to the
// same label.
*if_true = *if_false = *fall_through = materialize_true;
}
void FullCodeGenerator::AccumulatorValueContext::PrepareTest(
Label* materialize_true,
Label* materialize_false,
Label** if_true,
Label** if_false,
Label** fall_through) const {
*if_true = *fall_through = materialize_true;
*if_false = materialize_false;
}
void FullCodeGenerator::StackValueContext::PrepareTest(
Label* materialize_true,
Label* materialize_false,
Label** if_true,
Label** if_false,
Label** fall_through) const {
*if_true = *fall_through = materialize_true;
*if_false = materialize_false;
}
void FullCodeGenerator::TestContext::PrepareTest(
Label* materialize_true,
Label* materialize_false,
Label** if_true,
Label** if_false,
Label** fall_through) const {
*if_true = true_label_;
*if_false = false_label_;
*fall_through = fall_through_;
}
void FullCodeGenerator::DoTest(const TestContext* context) {
DoTest(context->condition(),
context->true_label(),
context->false_label(),
context->fall_through());
}
Get rid of static module allocation, do it in code. Modules now have their own local scope, represented by their own context. Module instance objects have an accessor for every export that forwards access to the respective slot from the module's context. (Exports that are modules themselves, however, are simple data properties.) All modules have a _hosting_ scope/context, which (currently) is the (innermost) enclosing global scope. To deal with recursion, nested modules are hosted by the same scope as global ones. For every (global or nested) module literal, the hosting context has an internal slot that points directly to the respective module context. This enables quick access to (statically resolved) module members by 2-dimensional access through the hosting context. For example, module A { let x; module B { let y; } } module C { let z; } allocates contexts as follows: [header| .A | .B | .C | A | C ] (global) | | | | | +-- [header| z ] (module) | | | +------- [header| y ] (module) | +------------ [header| x | B ] (module) Here, .A, .B, .C are the internal slots pointing to the hosted module contexts, whereas A, B, C hold the actual instance objects (note that every module context also points to the respective instance object through its extension slot in the header). To deal with arbitrary recursion and aliases between modules, they are created and initialized in several stages. Each stage applies to all modules in the hosting global scope, including nested ones. 1. Allocate: for each module _literal_, allocate the module contexts and respective instance object and wire them up. This happens in the PushModuleContext runtime function, as generated by AllocateModules (invoked by VisitDeclarations in the hosting scope). 2. Bind: for each module _declaration_ (i.e. literals as well as aliases), assign the respective instance object to respective local variables. This happens in VisitModuleDeclaration, and uses the instance objects created in the previous stage. For each module _literal_, this phase also constructs a module descriptor for the next stage. This happens in VisitModuleLiteral. 3. Populate: invoke the DeclareModules runtime function to populate each _instance_ object with accessors for it exports. This is generated by DeclareModules (invoked by VisitDeclarations in the hosting scope again), and uses the descriptors generated in the previous stage. 4. Initialize: execute the module bodies (and other code) in sequence. This happens by the separate statements generated for module bodies. To reenter the module scopes properly, the parser inserted ModuleStatements. R=mstarzinger@chromium.org,svenpanne@chromium.org BUG= Review URL: https://codereview.chromium.org/11093074 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@13033 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2012-11-22 10:25:22 +00:00
void FullCodeGenerator::AllocateModules(ZoneList<Declaration*>* declarations) {
ASSERT(scope_->is_global_scope());
for (int i = 0; i < declarations->length(); i++) {
ModuleDeclaration* declaration = declarations->at(i)->AsModuleDeclaration();
if (declaration != NULL) {
ModuleLiteral* module = declaration->module()->AsModuleLiteral();
if (module != NULL) {
Comment cmnt(masm_, "[ Link nested modules");
Scope* scope = module->body()->scope();
Interface* interface = scope->interface();
ASSERT(interface->IsModule() && interface->IsFrozen());
interface->Allocate(scope->module_var()->index());
// Set up module context.
ASSERT(scope->interface()->Index() >= 0);
__ Push(Smi::FromInt(scope->interface()->Index()));
__ Push(scope->GetScopeInfo());
__ CallRuntime(Runtime::kPushModuleContext, 2);
StoreToFrameField(StandardFrameConstants::kContextOffset,
context_register());
AllocateModules(scope->declarations());
// Pop module context.
LoadContextField(context_register(), Context::PREVIOUS_INDEX);
// Update local stack frame context field.
StoreToFrameField(StandardFrameConstants::kContextOffset,
context_register());
}
}
}
}
// Modules have their own local scope, represented by their own context.
// Module instance objects have an accessor for every export that forwards
// access to the respective slot from the module's context. (Exports that are
// modules themselves, however, are simple data properties.)
//
// All modules have a _hosting_ scope/context, which (currently) is the
// (innermost) enclosing global scope. To deal with recursion, nested modules
// are hosted by the same scope as global ones.
//
// For every (global or nested) module literal, the hosting context has an
// internal slot that points directly to the respective module context. This
// enables quick access to (statically resolved) module members by 2-dimensional
// access through the hosting context. For example,
//
// module A {
// let x;
// module B { let y; }
// }
// module C { let z; }
//
// allocates contexts as follows:
//
// [header| .A | .B | .C | A | C ] (global)
// | | |
// | | +-- [header| z ] (module)
// | |
// | +------- [header| y ] (module)
// |
// +------------ [header| x | B ] (module)
//
// Here, .A, .B, .C are the internal slots pointing to the hosted module
// contexts, whereas A, B, C hold the actual instance objects (note that every
// module context also points to the respective instance object through its
// extension slot in the header).
//
// To deal with arbitrary recursion and aliases between modules,
// they are created and initialized in several stages. Each stage applies to
// all modules in the hosting global scope, including nested ones.
//
// 1. Allocate: for each module _literal_, allocate the module contexts and
// respective instance object and wire them up. This happens in the
// PushModuleContext runtime function, as generated by AllocateModules
// (invoked by VisitDeclarations in the hosting scope).
//
// 2. Bind: for each module _declaration_ (i.e. literals as well as aliases),
// assign the respective instance object to respective local variables. This
// happens in VisitModuleDeclaration, and uses the instance objects created
// in the previous stage.
// For each module _literal_, this phase also constructs a module descriptor
// for the next stage. This happens in VisitModuleLiteral.
//
// 3. Populate: invoke the DeclareModules runtime function to populate each
// _instance_ object with accessors for it exports. This is generated by
// DeclareModules (invoked by VisitDeclarations in the hosting scope again),
// and uses the descriptors generated in the previous stage.
//
// 4. Initialize: execute the module bodies (and other code) in sequence. This
// happens by the separate statements generated for module bodies. To reenter
// the module scopes properly, the parser inserted ModuleStatements.
void FullCodeGenerator::VisitDeclarations(
ZoneList<Declaration*>* declarations) {
Get rid of static module allocation, do it in code. Modules now have their own local scope, represented by their own context. Module instance objects have an accessor for every export that forwards access to the respective slot from the module's context. (Exports that are modules themselves, however, are simple data properties.) All modules have a _hosting_ scope/context, which (currently) is the (innermost) enclosing global scope. To deal with recursion, nested modules are hosted by the same scope as global ones. For every (global or nested) module literal, the hosting context has an internal slot that points directly to the respective module context. This enables quick access to (statically resolved) module members by 2-dimensional access through the hosting context. For example, module A { let x; module B { let y; } } module C { let z; } allocates contexts as follows: [header| .A | .B | .C | A | C ] (global) | | | | | +-- [header| z ] (module) | | | +------- [header| y ] (module) | +------------ [header| x | B ] (module) Here, .A, .B, .C are the internal slots pointing to the hosted module contexts, whereas A, B, C hold the actual instance objects (note that every module context also points to the respective instance object through its extension slot in the header). To deal with arbitrary recursion and aliases between modules, they are created and initialized in several stages. Each stage applies to all modules in the hosting global scope, including nested ones. 1. Allocate: for each module _literal_, allocate the module contexts and respective instance object and wire them up. This happens in the PushModuleContext runtime function, as generated by AllocateModules (invoked by VisitDeclarations in the hosting scope). 2. Bind: for each module _declaration_ (i.e. literals as well as aliases), assign the respective instance object to respective local variables. This happens in VisitModuleDeclaration, and uses the instance objects created in the previous stage. For each module _literal_, this phase also constructs a module descriptor for the next stage. This happens in VisitModuleLiteral. 3. Populate: invoke the DeclareModules runtime function to populate each _instance_ object with accessors for it exports. This is generated by DeclareModules (invoked by VisitDeclarations in the hosting scope again), and uses the descriptors generated in the previous stage. 4. Initialize: execute the module bodies (and other code) in sequence. This happens by the separate statements generated for module bodies. To reenter the module scopes properly, the parser inserted ModuleStatements. R=mstarzinger@chromium.org,svenpanne@chromium.org BUG= Review URL: https://codereview.chromium.org/11093074 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@13033 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2012-11-22 10:25:22 +00:00
Handle<FixedArray> saved_modules = modules_;
int saved_module_index = module_index_;
ZoneList<Handle<Object> >* saved_globals = globals_;
ZoneList<Handle<Object> > inner_globals(10, zone());
globals_ = &inner_globals;
Get rid of static module allocation, do it in code. Modules now have their own local scope, represented by their own context. Module instance objects have an accessor for every export that forwards access to the respective slot from the module's context. (Exports that are modules themselves, however, are simple data properties.) All modules have a _hosting_ scope/context, which (currently) is the (innermost) enclosing global scope. To deal with recursion, nested modules are hosted by the same scope as global ones. For every (global or nested) module literal, the hosting context has an internal slot that points directly to the respective module context. This enables quick access to (statically resolved) module members by 2-dimensional access through the hosting context. For example, module A { let x; module B { let y; } } module C { let z; } allocates contexts as follows: [header| .A | .B | .C | A | C ] (global) | | | | | +-- [header| z ] (module) | | | +------- [header| y ] (module) | +------------ [header| x | B ] (module) Here, .A, .B, .C are the internal slots pointing to the hosted module contexts, whereas A, B, C hold the actual instance objects (note that every module context also points to the respective instance object through its extension slot in the header). To deal with arbitrary recursion and aliases between modules, they are created and initialized in several stages. Each stage applies to all modules in the hosting global scope, including nested ones. 1. Allocate: for each module _literal_, allocate the module contexts and respective instance object and wire them up. This happens in the PushModuleContext runtime function, as generated by AllocateModules (invoked by VisitDeclarations in the hosting scope). 2. Bind: for each module _declaration_ (i.e. literals as well as aliases), assign the respective instance object to respective local variables. This happens in VisitModuleDeclaration, and uses the instance objects created in the previous stage. For each module _literal_, this phase also constructs a module descriptor for the next stage. This happens in VisitModuleLiteral. 3. Populate: invoke the DeclareModules runtime function to populate each _instance_ object with accessors for it exports. This is generated by DeclareModules (invoked by VisitDeclarations in the hosting scope again), and uses the descriptors generated in the previous stage. 4. Initialize: execute the module bodies (and other code) in sequence. This happens by the separate statements generated for module bodies. To reenter the module scopes properly, the parser inserted ModuleStatements. R=mstarzinger@chromium.org,svenpanne@chromium.org BUG= Review URL: https://codereview.chromium.org/11093074 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@13033 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2012-11-22 10:25:22 +00:00
if (scope_->num_modules() != 0) {
// This is a scope hosting modules. Allocate a descriptor array to pass
// to the runtime for initialization.
Comment cmnt(masm_, "[ Allocate modules");
ASSERT(scope_->is_global_scope());
modules_ =
isolate()->factory()->NewFixedArray(scope_->num_modules(), TENURED);
module_index_ = 0;
// Generate code for allocating all modules, including nested ones.
// The allocated contexts are stored in internal variables in this scope.
AllocateModules(declarations);
}
AstVisitor::VisitDeclarations(declarations);
Get rid of static module allocation, do it in code. Modules now have their own local scope, represented by their own context. Module instance objects have an accessor for every export that forwards access to the respective slot from the module's context. (Exports that are modules themselves, however, are simple data properties.) All modules have a _hosting_ scope/context, which (currently) is the (innermost) enclosing global scope. To deal with recursion, nested modules are hosted by the same scope as global ones. For every (global or nested) module literal, the hosting context has an internal slot that points directly to the respective module context. This enables quick access to (statically resolved) module members by 2-dimensional access through the hosting context. For example, module A { let x; module B { let y; } } module C { let z; } allocates contexts as follows: [header| .A | .B | .C | A | C ] (global) | | | | | +-- [header| z ] (module) | | | +------- [header| y ] (module) | +------------ [header| x | B ] (module) Here, .A, .B, .C are the internal slots pointing to the hosted module contexts, whereas A, B, C hold the actual instance objects (note that every module context also points to the respective instance object through its extension slot in the header). To deal with arbitrary recursion and aliases between modules, they are created and initialized in several stages. Each stage applies to all modules in the hosting global scope, including nested ones. 1. Allocate: for each module _literal_, allocate the module contexts and respective instance object and wire them up. This happens in the PushModuleContext runtime function, as generated by AllocateModules (invoked by VisitDeclarations in the hosting scope). 2. Bind: for each module _declaration_ (i.e. literals as well as aliases), assign the respective instance object to respective local variables. This happens in VisitModuleDeclaration, and uses the instance objects created in the previous stage. For each module _literal_, this phase also constructs a module descriptor for the next stage. This happens in VisitModuleLiteral. 3. Populate: invoke the DeclareModules runtime function to populate each _instance_ object with accessors for it exports. This is generated by DeclareModules (invoked by VisitDeclarations in the hosting scope again), and uses the descriptors generated in the previous stage. 4. Initialize: execute the module bodies (and other code) in sequence. This happens by the separate statements generated for module bodies. To reenter the module scopes properly, the parser inserted ModuleStatements. R=mstarzinger@chromium.org,svenpanne@chromium.org BUG= Review URL: https://codereview.chromium.org/11093074 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@13033 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2012-11-22 10:25:22 +00:00
if (scope_->num_modules() != 0) {
// Initialize modules from descriptor array.
ASSERT(module_index_ == modules_->length());
DeclareModules(modules_);
modules_ = saved_modules;
module_index_ = saved_module_index;
}
if (!globals_->is_empty()) {
// Invoke the platform-dependent code generator to do the actual
Get rid of static module allocation, do it in code. Modules now have their own local scope, represented by their own context. Module instance objects have an accessor for every export that forwards access to the respective slot from the module's context. (Exports that are modules themselves, however, are simple data properties.) All modules have a _hosting_ scope/context, which (currently) is the (innermost) enclosing global scope. To deal with recursion, nested modules are hosted by the same scope as global ones. For every (global or nested) module literal, the hosting context has an internal slot that points directly to the respective module context. This enables quick access to (statically resolved) module members by 2-dimensional access through the hosting context. For example, module A { let x; module B { let y; } } module C { let z; } allocates contexts as follows: [header| .A | .B | .C | A | C ] (global) | | | | | +-- [header| z ] (module) | | | +------- [header| y ] (module) | +------------ [header| x | B ] (module) Here, .A, .B, .C are the internal slots pointing to the hosted module contexts, whereas A, B, C hold the actual instance objects (note that every module context also points to the respective instance object through its extension slot in the header). To deal with arbitrary recursion and aliases between modules, they are created and initialized in several stages. Each stage applies to all modules in the hosting global scope, including nested ones. 1. Allocate: for each module _literal_, allocate the module contexts and respective instance object and wire them up. This happens in the PushModuleContext runtime function, as generated by AllocateModules (invoked by VisitDeclarations in the hosting scope). 2. Bind: for each module _declaration_ (i.e. literals as well as aliases), assign the respective instance object to respective local variables. This happens in VisitModuleDeclaration, and uses the instance objects created in the previous stage. For each module _literal_, this phase also constructs a module descriptor for the next stage. This happens in VisitModuleLiteral. 3. Populate: invoke the DeclareModules runtime function to populate each _instance_ object with accessors for it exports. This is generated by DeclareModules (invoked by VisitDeclarations in the hosting scope again), and uses the descriptors generated in the previous stage. 4. Initialize: execute the module bodies (and other code) in sequence. This happens by the separate statements generated for module bodies. To reenter the module scopes properly, the parser inserted ModuleStatements. R=mstarzinger@chromium.org,svenpanne@chromium.org BUG= Review URL: https://codereview.chromium.org/11093074 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@13033 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2012-11-22 10:25:22 +00:00
// declaration of the global functions and variables.
Handle<FixedArray> array =
isolate()->factory()->NewFixedArray(globals_->length(), TENURED);
for (int i = 0; i < globals_->length(); ++i)
array->set(i, *globals_->at(i));
DeclareGlobals(array);
}
globals_ = saved_globals;
}
void FullCodeGenerator::VisitModuleLiteral(ModuleLiteral* module) {
Block* block = module->body();
Scope* saved_scope = scope();
scope_ = block->scope();
Get rid of static module allocation, do it in code. Modules now have their own local scope, represented by their own context. Module instance objects have an accessor for every export that forwards access to the respective slot from the module's context. (Exports that are modules themselves, however, are simple data properties.) All modules have a _hosting_ scope/context, which (currently) is the (innermost) enclosing global scope. To deal with recursion, nested modules are hosted by the same scope as global ones. For every (global or nested) module literal, the hosting context has an internal slot that points directly to the respective module context. This enables quick access to (statically resolved) module members by 2-dimensional access through the hosting context. For example, module A { let x; module B { let y; } } module C { let z; } allocates contexts as follows: [header| .A | .B | .C | A | C ] (global) | | | | | +-- [header| z ] (module) | | | +------- [header| y ] (module) | +------------ [header| x | B ] (module) Here, .A, .B, .C are the internal slots pointing to the hosted module contexts, whereas A, B, C hold the actual instance objects (note that every module context also points to the respective instance object through its extension slot in the header). To deal with arbitrary recursion and aliases between modules, they are created and initialized in several stages. Each stage applies to all modules in the hosting global scope, including nested ones. 1. Allocate: for each module _literal_, allocate the module contexts and respective instance object and wire them up. This happens in the PushModuleContext runtime function, as generated by AllocateModules (invoked by VisitDeclarations in the hosting scope). 2. Bind: for each module _declaration_ (i.e. literals as well as aliases), assign the respective instance object to respective local variables. This happens in VisitModuleDeclaration, and uses the instance objects created in the previous stage. For each module _literal_, this phase also constructs a module descriptor for the next stage. This happens in VisitModuleLiteral. 3. Populate: invoke the DeclareModules runtime function to populate each _instance_ object with accessors for it exports. This is generated by DeclareModules (invoked by VisitDeclarations in the hosting scope again), and uses the descriptors generated in the previous stage. 4. Initialize: execute the module bodies (and other code) in sequence. This happens by the separate statements generated for module bodies. To reenter the module scopes properly, the parser inserted ModuleStatements. R=mstarzinger@chromium.org,svenpanne@chromium.org BUG= Review URL: https://codereview.chromium.org/11093074 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@13033 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2012-11-22 10:25:22 +00:00
Interface* interface = scope_->interface();
Comment cmnt(masm_, "[ ModuleLiteral");
SetStatementPosition(block);
Get rid of static module allocation, do it in code. Modules now have their own local scope, represented by their own context. Module instance objects have an accessor for every export that forwards access to the respective slot from the module's context. (Exports that are modules themselves, however, are simple data properties.) All modules have a _hosting_ scope/context, which (currently) is the (innermost) enclosing global scope. To deal with recursion, nested modules are hosted by the same scope as global ones. For every (global or nested) module literal, the hosting context has an internal slot that points directly to the respective module context. This enables quick access to (statically resolved) module members by 2-dimensional access through the hosting context. For example, module A { let x; module B { let y; } } module C { let z; } allocates contexts as follows: [header| .A | .B | .C | A | C ] (global) | | | | | +-- [header| z ] (module) | | | +------- [header| y ] (module) | +------------ [header| x | B ] (module) Here, .A, .B, .C are the internal slots pointing to the hosted module contexts, whereas A, B, C hold the actual instance objects (note that every module context also points to the respective instance object through its extension slot in the header). To deal with arbitrary recursion and aliases between modules, they are created and initialized in several stages. Each stage applies to all modules in the hosting global scope, including nested ones. 1. Allocate: for each module _literal_, allocate the module contexts and respective instance object and wire them up. This happens in the PushModuleContext runtime function, as generated by AllocateModules (invoked by VisitDeclarations in the hosting scope). 2. Bind: for each module _declaration_ (i.e. literals as well as aliases), assign the respective instance object to respective local variables. This happens in VisitModuleDeclaration, and uses the instance objects created in the previous stage. For each module _literal_, this phase also constructs a module descriptor for the next stage. This happens in VisitModuleLiteral. 3. Populate: invoke the DeclareModules runtime function to populate each _instance_ object with accessors for it exports. This is generated by DeclareModules (invoked by VisitDeclarations in the hosting scope again), and uses the descriptors generated in the previous stage. 4. Initialize: execute the module bodies (and other code) in sequence. This happens by the separate statements generated for module bodies. To reenter the module scopes properly, the parser inserted ModuleStatements. R=mstarzinger@chromium.org,svenpanne@chromium.org BUG= Review URL: https://codereview.chromium.org/11093074 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@13033 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2012-11-22 10:25:22 +00:00
ASSERT(!modules_.is_null());
ASSERT(module_index_ < modules_->length());
int index = module_index_++;
// Set up module context.
Get rid of static module allocation, do it in code. Modules now have their own local scope, represented by their own context. Module instance objects have an accessor for every export that forwards access to the respective slot from the module's context. (Exports that are modules themselves, however, are simple data properties.) All modules have a _hosting_ scope/context, which (currently) is the (innermost) enclosing global scope. To deal with recursion, nested modules are hosted by the same scope as global ones. For every (global or nested) module literal, the hosting context has an internal slot that points directly to the respective module context. This enables quick access to (statically resolved) module members by 2-dimensional access through the hosting context. For example, module A { let x; module B { let y; } } module C { let z; } allocates contexts as follows: [header| .A | .B | .C | A | C ] (global) | | | | | +-- [header| z ] (module) | | | +------- [header| y ] (module) | +------------ [header| x | B ] (module) Here, .A, .B, .C are the internal slots pointing to the hosted module contexts, whereas A, B, C hold the actual instance objects (note that every module context also points to the respective instance object through its extension slot in the header). To deal with arbitrary recursion and aliases between modules, they are created and initialized in several stages. Each stage applies to all modules in the hosting global scope, including nested ones. 1. Allocate: for each module _literal_, allocate the module contexts and respective instance object and wire them up. This happens in the PushModuleContext runtime function, as generated by AllocateModules (invoked by VisitDeclarations in the hosting scope). 2. Bind: for each module _declaration_ (i.e. literals as well as aliases), assign the respective instance object to respective local variables. This happens in VisitModuleDeclaration, and uses the instance objects created in the previous stage. For each module _literal_, this phase also constructs a module descriptor for the next stage. This happens in VisitModuleLiteral. 3. Populate: invoke the DeclareModules runtime function to populate each _instance_ object with accessors for it exports. This is generated by DeclareModules (invoked by VisitDeclarations in the hosting scope again), and uses the descriptors generated in the previous stage. 4. Initialize: execute the module bodies (and other code) in sequence. This happens by the separate statements generated for module bodies. To reenter the module scopes properly, the parser inserted ModuleStatements. R=mstarzinger@chromium.org,svenpanne@chromium.org BUG= Review URL: https://codereview.chromium.org/11093074 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@13033 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2012-11-22 10:25:22 +00:00
ASSERT(interface->Index() >= 0);
__ Push(Smi::FromInt(interface->Index()));
__ Push(Smi::FromInt(0));
__ CallRuntime(Runtime::kPushModuleContext, 2);
StoreToFrameField(StandardFrameConstants::kContextOffset, context_register());
{
Comment cmnt(masm_, "[ Declarations");
VisitDeclarations(scope_->declarations());
}
Get rid of static module allocation, do it in code. Modules now have their own local scope, represented by their own context. Module instance objects have an accessor for every export that forwards access to the respective slot from the module's context. (Exports that are modules themselves, however, are simple data properties.) All modules have a _hosting_ scope/context, which (currently) is the (innermost) enclosing global scope. To deal with recursion, nested modules are hosted by the same scope as global ones. For every (global or nested) module literal, the hosting context has an internal slot that points directly to the respective module context. This enables quick access to (statically resolved) module members by 2-dimensional access through the hosting context. For example, module A { let x; module B { let y; } } module C { let z; } allocates contexts as follows: [header| .A | .B | .C | A | C ] (global) | | | | | +-- [header| z ] (module) | | | +------- [header| y ] (module) | +------------ [header| x | B ] (module) Here, .A, .B, .C are the internal slots pointing to the hosted module contexts, whereas A, B, C hold the actual instance objects (note that every module context also points to the respective instance object through its extension slot in the header). To deal with arbitrary recursion and aliases between modules, they are created and initialized in several stages. Each stage applies to all modules in the hosting global scope, including nested ones. 1. Allocate: for each module _literal_, allocate the module contexts and respective instance object and wire them up. This happens in the PushModuleContext runtime function, as generated by AllocateModules (invoked by VisitDeclarations in the hosting scope). 2. Bind: for each module _declaration_ (i.e. literals as well as aliases), assign the respective instance object to respective local variables. This happens in VisitModuleDeclaration, and uses the instance objects created in the previous stage. For each module _literal_, this phase also constructs a module descriptor for the next stage. This happens in VisitModuleLiteral. 3. Populate: invoke the DeclareModules runtime function to populate each _instance_ object with accessors for it exports. This is generated by DeclareModules (invoked by VisitDeclarations in the hosting scope again), and uses the descriptors generated in the previous stage. 4. Initialize: execute the module bodies (and other code) in sequence. This happens by the separate statements generated for module bodies. To reenter the module scopes properly, the parser inserted ModuleStatements. R=mstarzinger@chromium.org,svenpanne@chromium.org BUG= Review URL: https://codereview.chromium.org/11093074 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@13033 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2012-11-22 10:25:22 +00:00
// Populate the module description.
Handle<ModuleInfo> description =
ModuleInfo::Create(isolate(), interface, scope_);
modules_->set(index, *description);
scope_ = saved_scope;
// Pop module context.
LoadContextField(context_register(), Context::PREVIOUS_INDEX);
// Update local stack frame context field.
StoreToFrameField(StandardFrameConstants::kContextOffset, context_register());
}
void FullCodeGenerator::VisitModuleVariable(ModuleVariable* module) {
// Nothing to do.
// The instance object is resolved statically through the module's interface.
}
void FullCodeGenerator::VisitModulePath(ModulePath* module) {
// Nothing to do.
// The instance object is resolved statically through the module's interface.
}
Get rid of static module allocation, do it in code. Modules now have their own local scope, represented by their own context. Module instance objects have an accessor for every export that forwards access to the respective slot from the module's context. (Exports that are modules themselves, however, are simple data properties.) All modules have a _hosting_ scope/context, which (currently) is the (innermost) enclosing global scope. To deal with recursion, nested modules are hosted by the same scope as global ones. For every (global or nested) module literal, the hosting context has an internal slot that points directly to the respective module context. This enables quick access to (statically resolved) module members by 2-dimensional access through the hosting context. For example, module A { let x; module B { let y; } } module C { let z; } allocates contexts as follows: [header| .A | .B | .C | A | C ] (global) | | | | | +-- [header| z ] (module) | | | +------- [header| y ] (module) | +------------ [header| x | B ] (module) Here, .A, .B, .C are the internal slots pointing to the hosted module contexts, whereas A, B, C hold the actual instance objects (note that every module context also points to the respective instance object through its extension slot in the header). To deal with arbitrary recursion and aliases between modules, they are created and initialized in several stages. Each stage applies to all modules in the hosting global scope, including nested ones. 1. Allocate: for each module _literal_, allocate the module contexts and respective instance object and wire them up. This happens in the PushModuleContext runtime function, as generated by AllocateModules (invoked by VisitDeclarations in the hosting scope). 2. Bind: for each module _declaration_ (i.e. literals as well as aliases), assign the respective instance object to respective local variables. This happens in VisitModuleDeclaration, and uses the instance objects created in the previous stage. For each module _literal_, this phase also constructs a module descriptor for the next stage. This happens in VisitModuleLiteral. 3. Populate: invoke the DeclareModules runtime function to populate each _instance_ object with accessors for it exports. This is generated by DeclareModules (invoked by VisitDeclarations in the hosting scope again), and uses the descriptors generated in the previous stage. 4. Initialize: execute the module bodies (and other code) in sequence. This happens by the separate statements generated for module bodies. To reenter the module scopes properly, the parser inserted ModuleStatements. R=mstarzinger@chromium.org,svenpanne@chromium.org BUG= Review URL: https://codereview.chromium.org/11093074 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@13033 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2012-11-22 10:25:22 +00:00
void FullCodeGenerator::VisitModuleUrl(ModuleUrl* module) {
// TODO(rossberg): dummy allocation for now.
Scope* scope = module->body()->scope();
Interface* interface = scope_->interface();
ASSERT(interface->IsModule() && interface->IsFrozen());
ASSERT(!modules_.is_null());
ASSERT(module_index_ < modules_->length());
interface->Allocate(scope->module_var()->index());
int index = module_index_++;
Handle<ModuleInfo> description =
ModuleInfo::Create(isolate(), interface, scope_);
modules_->set(index, *description);
}
int FullCodeGenerator::DeclareGlobalsFlags() {
ASSERT(DeclareGlobalsLanguageMode::is_valid(language_mode()));
return DeclareGlobalsEvalFlag::encode(is_eval()) |
DeclareGlobalsNativeFlag::encode(is_native()) |
DeclareGlobalsLanguageMode::encode(language_mode());
}
void FullCodeGenerator::SetFunctionPosition(FunctionLiteral* fun) {
CodeGenerator::RecordPositions(masm_, fun->start_position());
}
void FullCodeGenerator::SetReturnPosition(FunctionLiteral* fun) {
CodeGenerator::RecordPositions(masm_, fun->end_position() - 1);
}
void FullCodeGenerator::SetStatementPosition(Statement* stmt) {
#ifdef ENABLE_DEBUGGER_SUPPORT
if (!isolate()->debugger()->IsDebuggerActive()) {
CodeGenerator::RecordPositions(masm_, stmt->statement_pos());
} else {
// Check if the statement will be breakable without adding a debug break
// slot.
BreakableStatementChecker checker;
checker.Check(stmt);
// Record the statement position right here if the statement is not
// breakable. For breakable statements the actual recording of the
// position will be postponed to the breakable code (typically an IC).
bool position_recorded = CodeGenerator::RecordPositions(
masm_, stmt->statement_pos(), !checker.is_breakable());
// If the position recording did record a new position generate a debug
// break slot to make the statement breakable.
if (position_recorded) {
Debug::GenerateSlot(masm_);
}
}
#else
CodeGenerator::RecordPositions(masm_, stmt->statement_pos());
#endif
}
void FullCodeGenerator::SetExpressionPosition(Expression* expr, int pos) {
#ifdef ENABLE_DEBUGGER_SUPPORT
if (!isolate()->debugger()->IsDebuggerActive()) {
CodeGenerator::RecordPositions(masm_, pos);
} else {
// Check if the expression will be breakable without adding a debug break
// slot.
BreakableStatementChecker checker;
checker.Check(expr);
// Record a statement position right here if the expression is not
// breakable. For breakable expressions the actual recording of the
// position will be postponed to the breakable code (typically an IC).
// NOTE this will record a statement position for something which might
// not be a statement. As stepping in the debugger will only stop at
// statement positions this is used for e.g. the condition expression of
// a do while loop.
bool position_recorded = CodeGenerator::RecordPositions(
masm_, pos, !checker.is_breakable());
// If the position recording did record a new position generate a debug
// break slot to make the statement breakable.
if (position_recorded) {
Debug::GenerateSlot(masm_);
}
}
#else
CodeGenerator::RecordPositions(masm_, pos);
#endif
}
void FullCodeGenerator::SetStatementPosition(int pos) {
CodeGenerator::RecordPositions(masm_, pos);
}
void FullCodeGenerator::SetSourcePosition(int pos) {
if (pos != RelocInfo::kNoPosition) {
masm_->positions_recorder()->RecordPosition(pos);
}
}
Initial infrastructure for fast compilation of top-level code. The fast code generator is optimized for compilation time and code size. Currently it is only implemented on IA32. It is potentially triggered for any code in the global scope (including code eval'd in the global scope). It performs a syntactic check and chooses to compile in fast mode if the AST contains only supported constructs and matches some other constraints. Initially supported constructs are * ExpressionStatement, * ReturnStatement, * VariableProxy (variable references) to parameters and stack-allocated locals, * Assignment with lhs a parameter or stack-allocated local, and * Literal This allows compilation of literals at the top level and not much else. All intermediate values are allocated to temporaries and the stack is used for all temporaries. The extra memory traffic is a known issue. The code generated for 'true' is: 0 push ebp 1 mov ebp,esp 3 push esi 4 push edi 5 push 0xf5cca135 ;; object: 0xf5cca135 <undefined> 10 cmp esp,[0x8277efc] 16 jnc 27 (0xf5cbbb1b) 22 call 0xf5cac960 ;; code: STUB, StackCheck, minor: 0 27 push 0xf5cca161 ;; object: 0xf5cca161 <true> 32 mov eax,[esp] 35 mov [ebp+0xf4],eax 38 pop eax 39 mov eax,[ebp+0xf4] 42 mov esp,ebp ;; js return 44 pop ebp 45 ret 0x4 48 mov eax,0xf5cca135 ;; object: 0xf5cca135 <undefined> 53 mov esp,ebp ;; js return 55 pop ebp 56 ret 0x4 Review URL: http://codereview.chromium.org/273050 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@3067 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-10-14 19:30:50 +00:00
// Lookup table for code generators for special runtime calls which are
// generated inline.
#define INLINE_FUNCTION_GENERATOR_ADDRESS(Name, argc, ressize) \
&FullCodeGenerator::Emit##Name,
const FullCodeGenerator::InlineFunctionGenerator
FullCodeGenerator::kInlineFunctionGenerators[] = {
INLINE_FUNCTION_LIST(INLINE_FUNCTION_GENERATOR_ADDRESS)
INLINE_RUNTIME_FUNCTION_LIST(INLINE_FUNCTION_GENERATOR_ADDRESS)
};
#undef INLINE_FUNCTION_GENERATOR_ADDRESS
FullCodeGenerator::InlineFunctionGenerator
FullCodeGenerator::FindInlineFunctionGenerator(Runtime::FunctionId id) {
int lookup_index =
static_cast<int>(id) - static_cast<int>(Runtime::kFirstInlineFunction);
ASSERT(lookup_index >= 0);
ASSERT(static_cast<size_t>(lookup_index) <
ARRAY_SIZE(kInlineFunctionGenerators));
return kInlineFunctionGenerators[lookup_index];
}
void FullCodeGenerator::EmitInlineRuntimeCall(CallRuntime* expr) {
const Runtime::Function* function = expr->function();
ASSERT(function != NULL);
ASSERT(function->intrinsic_type == Runtime::INLINE);
InlineFunctionGenerator generator =
FindInlineFunctionGenerator(function->function_id);
((*this).*(generator))(expr);
}
void FullCodeGenerator::VisitBinaryOperation(BinaryOperation* expr) {
switch (expr->op()) {
case Token::COMMA:
return VisitComma(expr);
case Token::OR:
case Token::AND:
return VisitLogicalExpression(expr);
default:
return VisitArithmeticExpression(expr);
}
}
void FullCodeGenerator::VisitInDuplicateContext(Expression* expr) {
if (context()->IsEffect()) {
VisitForEffect(expr);
} else if (context()->IsAccumulatorValue()) {
VisitForAccumulatorValue(expr);
} else if (context()->IsStackValue()) {
VisitForStackValue(expr);
} else if (context()->IsTest()) {
const TestContext* test = TestContext::cast(context());
VisitForControl(expr, test->true_label(), test->false_label(),
test->fall_through());
}
}
void FullCodeGenerator::VisitComma(BinaryOperation* expr) {
Comment cmnt(masm_, "[ Comma");
VisitForEffect(expr->left());
VisitInDuplicateContext(expr->right());
}
void FullCodeGenerator::VisitLogicalExpression(BinaryOperation* expr) {
bool is_logical_and = expr->op() == Token::AND;
Comment cmnt(masm_, is_logical_and ? "[ Logical AND" : "[ Logical OR");
Expression* left = expr->left();
Expression* right = expr->right();
BailoutId right_id = expr->RightId();
Label done;
if (context()->IsTest()) {
Label eval_right;
const TestContext* test = TestContext::cast(context());
if (is_logical_and) {
VisitForControl(left, &eval_right, test->false_label(), &eval_right);
} else {
VisitForControl(left, test->true_label(), &eval_right, &eval_right);
}
PrepareForBailoutForId(right_id, NO_REGISTERS);
__ bind(&eval_right);
} else if (context()->IsAccumulatorValue()) {
VisitForAccumulatorValue(left);
// We want the value in the accumulator for the test, and on the stack in
// case we need it.
__ push(result_register());
Label discard, restore;
if (is_logical_and) {
DoTest(left, &discard, &restore, &restore);
} else {
DoTest(left, &restore, &discard, &restore);
}
__ bind(&restore);
__ pop(result_register());
__ jmp(&done);
__ bind(&discard);
__ Drop(1);
PrepareForBailoutForId(right_id, NO_REGISTERS);
} else if (context()->IsStackValue()) {
VisitForAccumulatorValue(left);
// We want the value in the accumulator for the test, and on the stack in
// case we need it.
__ push(result_register());
Label discard;
if (is_logical_and) {
DoTest(left, &discard, &done, &discard);
} else {
DoTest(left, &done, &discard, &discard);
}
__ bind(&discard);
__ Drop(1);
PrepareForBailoutForId(right_id, NO_REGISTERS);
} else {
ASSERT(context()->IsEffect());
Label eval_right;
if (is_logical_and) {
VisitForControl(left, &eval_right, &done, &eval_right);
} else {
VisitForControl(left, &done, &eval_right, &eval_right);
}
PrepareForBailoutForId(right_id, NO_REGISTERS);
__ bind(&eval_right);
}
VisitInDuplicateContext(right);
__ bind(&done);
}
void FullCodeGenerator::VisitArithmeticExpression(BinaryOperation* expr) {
Token::Value op = expr->op();
Comment cmnt(masm_, "[ ArithmeticExpression");
Expression* left = expr->left();
Expression* right = expr->right();
OverwriteMode mode =
left->ResultOverwriteAllowed()
? OVERWRITE_LEFT
: (right->ResultOverwriteAllowed() ? OVERWRITE_RIGHT : NO_OVERWRITE);
VisitForStackValue(left);
VisitForAccumulatorValue(right);
SetSourcePosition(expr->position());
if (ShouldInlineSmiCase(op)) {
EmitInlineSmiBinaryOp(expr, op, mode, left, right);
} else {
EmitBinaryOp(expr, op, mode);
}
}
void FullCodeGenerator::VisitBlock(Block* stmt) {
Comment cmnt(masm_, "[ Block");
NestedBlock nested_block(this, stmt);
SetStatementPosition(stmt);
Scope* saved_scope = scope();
// Push a block context when entering a block with block scoped variables.
if (stmt->scope() != NULL) {
scope_ = stmt->scope();
Get rid of static module allocation, do it in code. Modules now have their own local scope, represented by their own context. Module instance objects have an accessor for every export that forwards access to the respective slot from the module's context. (Exports that are modules themselves, however, are simple data properties.) All modules have a _hosting_ scope/context, which (currently) is the (innermost) enclosing global scope. To deal with recursion, nested modules are hosted by the same scope as global ones. For every (global or nested) module literal, the hosting context has an internal slot that points directly to the respective module context. This enables quick access to (statically resolved) module members by 2-dimensional access through the hosting context. For example, module A { let x; module B { let y; } } module C { let z; } allocates contexts as follows: [header| .A | .B | .C | A | C ] (global) | | | | | +-- [header| z ] (module) | | | +------- [header| y ] (module) | +------------ [header| x | B ] (module) Here, .A, .B, .C are the internal slots pointing to the hosted module contexts, whereas A, B, C hold the actual instance objects (note that every module context also points to the respective instance object through its extension slot in the header). To deal with arbitrary recursion and aliases between modules, they are created and initialized in several stages. Each stage applies to all modules in the hosting global scope, including nested ones. 1. Allocate: for each module _literal_, allocate the module contexts and respective instance object and wire them up. This happens in the PushModuleContext runtime function, as generated by AllocateModules (invoked by VisitDeclarations in the hosting scope). 2. Bind: for each module _declaration_ (i.e. literals as well as aliases), assign the respective instance object to respective local variables. This happens in VisitModuleDeclaration, and uses the instance objects created in the previous stage. For each module _literal_, this phase also constructs a module descriptor for the next stage. This happens in VisitModuleLiteral. 3. Populate: invoke the DeclareModules runtime function to populate each _instance_ object with accessors for it exports. This is generated by DeclareModules (invoked by VisitDeclarations in the hosting scope again), and uses the descriptors generated in the previous stage. 4. Initialize: execute the module bodies (and other code) in sequence. This happens by the separate statements generated for module bodies. To reenter the module scopes properly, the parser inserted ModuleStatements. R=mstarzinger@chromium.org,svenpanne@chromium.org BUG= Review URL: https://codereview.chromium.org/11093074 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@13033 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2012-11-22 10:25:22 +00:00
ASSERT(!scope_->is_module_scope());
{ Comment cmnt(masm_, "[ Extend block context");
Handle<ScopeInfo> scope_info = scope_->GetScopeInfo();
int heap_slots = scope_info->ContextLength() - Context::MIN_CONTEXT_SLOTS;
__ Push(scope_info);
PushFunctionArgumentForContextAllocation();
if (heap_slots <= FastNewBlockContextStub::kMaximumSlots) {
FastNewBlockContextStub stub(heap_slots);
__ CallStub(&stub);
} else {
__ CallRuntime(Runtime::kPushBlockContext, 2);
}
Get rid of static module allocation, do it in code. Modules now have their own local scope, represented by their own context. Module instance objects have an accessor for every export that forwards access to the respective slot from the module's context. (Exports that are modules themselves, however, are simple data properties.) All modules have a _hosting_ scope/context, which (currently) is the (innermost) enclosing global scope. To deal with recursion, nested modules are hosted by the same scope as global ones. For every (global or nested) module literal, the hosting context has an internal slot that points directly to the respective module context. This enables quick access to (statically resolved) module members by 2-dimensional access through the hosting context. For example, module A { let x; module B { let y; } } module C { let z; } allocates contexts as follows: [header| .A | .B | .C | A | C ] (global) | | | | | +-- [header| z ] (module) | | | +------- [header| y ] (module) | +------------ [header| x | B ] (module) Here, .A, .B, .C are the internal slots pointing to the hosted module contexts, whereas A, B, C hold the actual instance objects (note that every module context also points to the respective instance object through its extension slot in the header). To deal with arbitrary recursion and aliases between modules, they are created and initialized in several stages. Each stage applies to all modules in the hosting global scope, including nested ones. 1. Allocate: for each module _literal_, allocate the module contexts and respective instance object and wire them up. This happens in the PushModuleContext runtime function, as generated by AllocateModules (invoked by VisitDeclarations in the hosting scope). 2. Bind: for each module _declaration_ (i.e. literals as well as aliases), assign the respective instance object to respective local variables. This happens in VisitModuleDeclaration, and uses the instance objects created in the previous stage. For each module _literal_, this phase also constructs a module descriptor for the next stage. This happens in VisitModuleLiteral. 3. Populate: invoke the DeclareModules runtime function to populate each _instance_ object with accessors for it exports. This is generated by DeclareModules (invoked by VisitDeclarations in the hosting scope again), and uses the descriptors generated in the previous stage. 4. Initialize: execute the module bodies (and other code) in sequence. This happens by the separate statements generated for module bodies. To reenter the module scopes properly, the parser inserted ModuleStatements. R=mstarzinger@chromium.org,svenpanne@chromium.org BUG= Review URL: https://codereview.chromium.org/11093074 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@13033 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2012-11-22 10:25:22 +00:00
// Replace the context stored in the frame.
StoreToFrameField(StandardFrameConstants::kContextOffset,
context_register());
}
{ Comment cmnt(masm_, "[ Declarations");
VisitDeclarations(scope_->declarations());
}
}
Get rid of static module allocation, do it in code. Modules now have their own local scope, represented by their own context. Module instance objects have an accessor for every export that forwards access to the respective slot from the module's context. (Exports that are modules themselves, however, are simple data properties.) All modules have a _hosting_ scope/context, which (currently) is the (innermost) enclosing global scope. To deal with recursion, nested modules are hosted by the same scope as global ones. For every (global or nested) module literal, the hosting context has an internal slot that points directly to the respective module context. This enables quick access to (statically resolved) module members by 2-dimensional access through the hosting context. For example, module A { let x; module B { let y; } } module C { let z; } allocates contexts as follows: [header| .A | .B | .C | A | C ] (global) | | | | | +-- [header| z ] (module) | | | +------- [header| y ] (module) | +------------ [header| x | B ] (module) Here, .A, .B, .C are the internal slots pointing to the hosted module contexts, whereas A, B, C hold the actual instance objects (note that every module context also points to the respective instance object through its extension slot in the header). To deal with arbitrary recursion and aliases between modules, they are created and initialized in several stages. Each stage applies to all modules in the hosting global scope, including nested ones. 1. Allocate: for each module _literal_, allocate the module contexts and respective instance object and wire them up. This happens in the PushModuleContext runtime function, as generated by AllocateModules (invoked by VisitDeclarations in the hosting scope). 2. Bind: for each module _declaration_ (i.e. literals as well as aliases), assign the respective instance object to respective local variables. This happens in VisitModuleDeclaration, and uses the instance objects created in the previous stage. For each module _literal_, this phase also constructs a module descriptor for the next stage. This happens in VisitModuleLiteral. 3. Populate: invoke the DeclareModules runtime function to populate each _instance_ object with accessors for it exports. This is generated by DeclareModules (invoked by VisitDeclarations in the hosting scope again), and uses the descriptors generated in the previous stage. 4. Initialize: execute the module bodies (and other code) in sequence. This happens by the separate statements generated for module bodies. To reenter the module scopes properly, the parser inserted ModuleStatements. R=mstarzinger@chromium.org,svenpanne@chromium.org BUG= Review URL: https://codereview.chromium.org/11093074 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@13033 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2012-11-22 10:25:22 +00:00
PrepareForBailoutForId(stmt->EntryId(), NO_REGISTERS);
VisitStatements(stmt->statements());
scope_ = saved_scope;
__ bind(nested_block.break_label());
PrepareForBailoutForId(stmt->ExitId(), NO_REGISTERS);
// Pop block context if necessary.
if (stmt->scope() != NULL) {
LoadContextField(context_register(), Context::PREVIOUS_INDEX);
// Update local stack frame context field.
StoreToFrameField(StandardFrameConstants::kContextOffset,
context_register());
}
}
Get rid of static module allocation, do it in code. Modules now have their own local scope, represented by their own context. Module instance objects have an accessor for every export that forwards access to the respective slot from the module's context. (Exports that are modules themselves, however, are simple data properties.) All modules have a _hosting_ scope/context, which (currently) is the (innermost) enclosing global scope. To deal with recursion, nested modules are hosted by the same scope as global ones. For every (global or nested) module literal, the hosting context has an internal slot that points directly to the respective module context. This enables quick access to (statically resolved) module members by 2-dimensional access through the hosting context. For example, module A { let x; module B { let y; } } module C { let z; } allocates contexts as follows: [header| .A | .B | .C | A | C ] (global) | | | | | +-- [header| z ] (module) | | | +------- [header| y ] (module) | +------------ [header| x | B ] (module) Here, .A, .B, .C are the internal slots pointing to the hosted module contexts, whereas A, B, C hold the actual instance objects (note that every module context also points to the respective instance object through its extension slot in the header). To deal with arbitrary recursion and aliases between modules, they are created and initialized in several stages. Each stage applies to all modules in the hosting global scope, including nested ones. 1. Allocate: for each module _literal_, allocate the module contexts and respective instance object and wire them up. This happens in the PushModuleContext runtime function, as generated by AllocateModules (invoked by VisitDeclarations in the hosting scope). 2. Bind: for each module _declaration_ (i.e. literals as well as aliases), assign the respective instance object to respective local variables. This happens in VisitModuleDeclaration, and uses the instance objects created in the previous stage. For each module _literal_, this phase also constructs a module descriptor for the next stage. This happens in VisitModuleLiteral. 3. Populate: invoke the DeclareModules runtime function to populate each _instance_ object with accessors for it exports. This is generated by DeclareModules (invoked by VisitDeclarations in the hosting scope again), and uses the descriptors generated in the previous stage. 4. Initialize: execute the module bodies (and other code) in sequence. This happens by the separate statements generated for module bodies. To reenter the module scopes properly, the parser inserted ModuleStatements. R=mstarzinger@chromium.org,svenpanne@chromium.org BUG= Review URL: https://codereview.chromium.org/11093074 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@13033 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2012-11-22 10:25:22 +00:00
void FullCodeGenerator::VisitModuleStatement(ModuleStatement* stmt) {
Comment cmnt(masm_, "[ Module context");
__ Push(Smi::FromInt(stmt->proxy()->interface()->Index()));
__ Push(Smi::FromInt(0));
__ CallRuntime(Runtime::kPushModuleContext, 2);
StoreToFrameField(
StandardFrameConstants::kContextOffset, context_register());
Scope* saved_scope = scope_;
scope_ = stmt->body()->scope();
VisitStatements(stmt->body()->statements());
scope_ = saved_scope;
LoadContextField(context_register(), Context::PREVIOUS_INDEX);
// Update local stack frame context field.
StoreToFrameField(StandardFrameConstants::kContextOffset,
context_register());
}
void FullCodeGenerator::VisitExpressionStatement(ExpressionStatement* stmt) {
Comment cmnt(masm_, "[ ExpressionStatement");
SetStatementPosition(stmt);
VisitForEffect(stmt->expression());
}
void FullCodeGenerator::VisitEmptyStatement(EmptyStatement* stmt) {
Comment cmnt(masm_, "[ EmptyStatement");
SetStatementPosition(stmt);
Initial infrastructure for fast compilation of top-level code. The fast code generator is optimized for compilation time and code size. Currently it is only implemented on IA32. It is potentially triggered for any code in the global scope (including code eval'd in the global scope). It performs a syntactic check and chooses to compile in fast mode if the AST contains only supported constructs and matches some other constraints. Initially supported constructs are * ExpressionStatement, * ReturnStatement, * VariableProxy (variable references) to parameters and stack-allocated locals, * Assignment with lhs a parameter or stack-allocated local, and * Literal This allows compilation of literals at the top level and not much else. All intermediate values are allocated to temporaries and the stack is used for all temporaries. The extra memory traffic is a known issue. The code generated for 'true' is: 0 push ebp 1 mov ebp,esp 3 push esi 4 push edi 5 push 0xf5cca135 ;; object: 0xf5cca135 <undefined> 10 cmp esp,[0x8277efc] 16 jnc 27 (0xf5cbbb1b) 22 call 0xf5cac960 ;; code: STUB, StackCheck, minor: 0 27 push 0xf5cca161 ;; object: 0xf5cca161 <true> 32 mov eax,[esp] 35 mov [ebp+0xf4],eax 38 pop eax 39 mov eax,[ebp+0xf4] 42 mov esp,ebp ;; js return 44 pop ebp 45 ret 0x4 48 mov eax,0xf5cca135 ;; object: 0xf5cca135 <undefined> 53 mov esp,ebp ;; js return 55 pop ebp 56 ret 0x4 Review URL: http://codereview.chromium.org/273050 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@3067 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-10-14 19:30:50 +00:00
}
void FullCodeGenerator::VisitIfStatement(IfStatement* stmt) {
Comment cmnt(masm_, "[ IfStatement");
SetStatementPosition(stmt);
Label then_part, else_part, done;
if (stmt->HasElseStatement()) {
VisitForControl(stmt->condition(), &then_part, &else_part, &then_part);
PrepareForBailoutForId(stmt->ThenId(), NO_REGISTERS);
__ bind(&then_part);
Visit(stmt->then_statement());
__ jmp(&done);
PrepareForBailoutForId(stmt->ElseId(), NO_REGISTERS);
__ bind(&else_part);
Visit(stmt->else_statement());
} else {
VisitForControl(stmt->condition(), &then_part, &done, &then_part);
PrepareForBailoutForId(stmt->ThenId(), NO_REGISTERS);
__ bind(&then_part);
Visit(stmt->then_statement());
PrepareForBailoutForId(stmt->ElseId(), NO_REGISTERS);
}
__ bind(&done);
PrepareForBailoutForId(stmt->IfId(), NO_REGISTERS);
Initial infrastructure for fast compilation of top-level code. The fast code generator is optimized for compilation time and code size. Currently it is only implemented on IA32. It is potentially triggered for any code in the global scope (including code eval'd in the global scope). It performs a syntactic check and chooses to compile in fast mode if the AST contains only supported constructs and matches some other constraints. Initially supported constructs are * ExpressionStatement, * ReturnStatement, * VariableProxy (variable references) to parameters and stack-allocated locals, * Assignment with lhs a parameter or stack-allocated local, and * Literal This allows compilation of literals at the top level and not much else. All intermediate values are allocated to temporaries and the stack is used for all temporaries. The extra memory traffic is a known issue. The code generated for 'true' is: 0 push ebp 1 mov ebp,esp 3 push esi 4 push edi 5 push 0xf5cca135 ;; object: 0xf5cca135 <undefined> 10 cmp esp,[0x8277efc] 16 jnc 27 (0xf5cbbb1b) 22 call 0xf5cac960 ;; code: STUB, StackCheck, minor: 0 27 push 0xf5cca161 ;; object: 0xf5cca161 <true> 32 mov eax,[esp] 35 mov [ebp+0xf4],eax 38 pop eax 39 mov eax,[ebp+0xf4] 42 mov esp,ebp ;; js return 44 pop ebp 45 ret 0x4 48 mov eax,0xf5cca135 ;; object: 0xf5cca135 <undefined> 53 mov esp,ebp ;; js return 55 pop ebp 56 ret 0x4 Review URL: http://codereview.chromium.org/273050 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@3067 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-10-14 19:30:50 +00:00
}
void FullCodeGenerator::VisitContinueStatement(ContinueStatement* stmt) {
Comment cmnt(masm_, "[ ContinueStatement");
SetStatementPosition(stmt);
NestedStatement* current = nesting_stack_;
int stack_depth = 0;
int context_length = 0;
// When continuing, we clobber the unpredictable value in the accumulator
// with one that's safe for GC. If we hit an exit from the try block of
// try...finally on our way out, we will unconditionally preserve the
// accumulator on the stack.
ClearAccumulator();
while (!current->IsContinueTarget(stmt->target())) {
current = current->Exit(&stack_depth, &context_length);
}
__ Drop(stack_depth);
if (context_length > 0) {
while (context_length > 0) {
LoadContextField(context_register(), Context::PREVIOUS_INDEX);
--context_length;
}
StoreToFrameField(StandardFrameConstants::kContextOffset,
context_register());
}
__ jmp(current->AsIteration()->continue_label());
Initial infrastructure for fast compilation of top-level code. The fast code generator is optimized for compilation time and code size. Currently it is only implemented on IA32. It is potentially triggered for any code in the global scope (including code eval'd in the global scope). It performs a syntactic check and chooses to compile in fast mode if the AST contains only supported constructs and matches some other constraints. Initially supported constructs are * ExpressionStatement, * ReturnStatement, * VariableProxy (variable references) to parameters and stack-allocated locals, * Assignment with lhs a parameter or stack-allocated local, and * Literal This allows compilation of literals at the top level and not much else. All intermediate values are allocated to temporaries and the stack is used for all temporaries. The extra memory traffic is a known issue. The code generated for 'true' is: 0 push ebp 1 mov ebp,esp 3 push esi 4 push edi 5 push 0xf5cca135 ;; object: 0xf5cca135 <undefined> 10 cmp esp,[0x8277efc] 16 jnc 27 (0xf5cbbb1b) 22 call 0xf5cac960 ;; code: STUB, StackCheck, minor: 0 27 push 0xf5cca161 ;; object: 0xf5cca161 <true> 32 mov eax,[esp] 35 mov [ebp+0xf4],eax 38 pop eax 39 mov eax,[ebp+0xf4] 42 mov esp,ebp ;; js return 44 pop ebp 45 ret 0x4 48 mov eax,0xf5cca135 ;; object: 0xf5cca135 <undefined> 53 mov esp,ebp ;; js return 55 pop ebp 56 ret 0x4 Review URL: http://codereview.chromium.org/273050 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@3067 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-10-14 19:30:50 +00:00
}
void FullCodeGenerator::VisitBreakStatement(BreakStatement* stmt) {
Comment cmnt(masm_, "[ BreakStatement");
SetStatementPosition(stmt);
NestedStatement* current = nesting_stack_;
int stack_depth = 0;
int context_length = 0;
// When breaking, we clobber the unpredictable value in the accumulator
// with one that's safe for GC. If we hit an exit from the try block of
// try...finally on our way out, we will unconditionally preserve the
// accumulator on the stack.
ClearAccumulator();
while (!current->IsBreakTarget(stmt->target())) {
current = current->Exit(&stack_depth, &context_length);
}
__ Drop(stack_depth);
if (context_length > 0) {
while (context_length > 0) {
LoadContextField(context_register(), Context::PREVIOUS_INDEX);
--context_length;
}
StoreToFrameField(StandardFrameConstants::kContextOffset,
context_register());
}
__ jmp(current->AsBreakable()->break_label());
Initial infrastructure for fast compilation of top-level code. The fast code generator is optimized for compilation time and code size. Currently it is only implemented on IA32. It is potentially triggered for any code in the global scope (including code eval'd in the global scope). It performs a syntactic check and chooses to compile in fast mode if the AST contains only supported constructs and matches some other constraints. Initially supported constructs are * ExpressionStatement, * ReturnStatement, * VariableProxy (variable references) to parameters and stack-allocated locals, * Assignment with lhs a parameter or stack-allocated local, and * Literal This allows compilation of literals at the top level and not much else. All intermediate values are allocated to temporaries and the stack is used for all temporaries. The extra memory traffic is a known issue. The code generated for 'true' is: 0 push ebp 1 mov ebp,esp 3 push esi 4 push edi 5 push 0xf5cca135 ;; object: 0xf5cca135 <undefined> 10 cmp esp,[0x8277efc] 16 jnc 27 (0xf5cbbb1b) 22 call 0xf5cac960 ;; code: STUB, StackCheck, minor: 0 27 push 0xf5cca161 ;; object: 0xf5cca161 <true> 32 mov eax,[esp] 35 mov [ebp+0xf4],eax 38 pop eax 39 mov eax,[ebp+0xf4] 42 mov esp,ebp ;; js return 44 pop ebp 45 ret 0x4 48 mov eax,0xf5cca135 ;; object: 0xf5cca135 <undefined> 53 mov esp,ebp ;; js return 55 pop ebp 56 ret 0x4 Review URL: http://codereview.chromium.org/273050 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@3067 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-10-14 19:30:50 +00:00
}
void FullCodeGenerator::VisitReturnStatement(ReturnStatement* stmt) {
Comment cmnt(masm_, "[ ReturnStatement");
SetStatementPosition(stmt);
Expression* expr = stmt->expression();
VisitForAccumulatorValue(expr);
// Exit all nested statements.
NestedStatement* current = nesting_stack_;
int stack_depth = 0;
int context_length = 0;
while (current != NULL) {
current = current->Exit(&stack_depth, &context_length);
}
__ Drop(stack_depth);
EmitReturnSequence();
}
void FullCodeGenerator::VisitWithStatement(WithStatement* stmt) {
Comment cmnt(masm_, "[ WithStatement");
SetStatementPosition(stmt);
VisitForStackValue(stmt->expression());
PushFunctionArgumentForContextAllocation();
__ CallRuntime(Runtime::kPushWithContext, 2);
StoreToFrameField(StandardFrameConstants::kContextOffset, context_register());
{ WithOrCatch body(this);
Visit(stmt->statement());
}
// Pop context.
LoadContextField(context_register(), Context::PREVIOUS_INDEX);
// Update local stack frame context field.
StoreToFrameField(StandardFrameConstants::kContextOffset, context_register());
Initial infrastructure for fast compilation of top-level code. The fast code generator is optimized for compilation time and code size. Currently it is only implemented on IA32. It is potentially triggered for any code in the global scope (including code eval'd in the global scope). It performs a syntactic check and chooses to compile in fast mode if the AST contains only supported constructs and matches some other constraints. Initially supported constructs are * ExpressionStatement, * ReturnStatement, * VariableProxy (variable references) to parameters and stack-allocated locals, * Assignment with lhs a parameter or stack-allocated local, and * Literal This allows compilation of literals at the top level and not much else. All intermediate values are allocated to temporaries and the stack is used for all temporaries. The extra memory traffic is a known issue. The code generated for 'true' is: 0 push ebp 1 mov ebp,esp 3 push esi 4 push edi 5 push 0xf5cca135 ;; object: 0xf5cca135 <undefined> 10 cmp esp,[0x8277efc] 16 jnc 27 (0xf5cbbb1b) 22 call 0xf5cac960 ;; code: STUB, StackCheck, minor: 0 27 push 0xf5cca161 ;; object: 0xf5cca161 <true> 32 mov eax,[esp] 35 mov [ebp+0xf4],eax 38 pop eax 39 mov eax,[ebp+0xf4] 42 mov esp,ebp ;; js return 44 pop ebp 45 ret 0x4 48 mov eax,0xf5cca135 ;; object: 0xf5cca135 <undefined> 53 mov esp,ebp ;; js return 55 pop ebp 56 ret 0x4 Review URL: http://codereview.chromium.org/273050 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@3067 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-10-14 19:30:50 +00:00
}
void FullCodeGenerator::VisitDoWhileStatement(DoWhileStatement* stmt) {
Comment cmnt(masm_, "[ DoWhileStatement");
SetStatementPosition(stmt);
Label body, stack_check;
Iteration loop_statement(this, stmt);
increment_loop_depth();
__ bind(&body);
Visit(stmt->body());
// Record the position of the do while condition and make sure it is
// possible to break on the condition.
__ bind(loop_statement.continue_label());
PrepareForBailoutForId(stmt->ContinueId(), NO_REGISTERS);
SetExpressionPosition(stmt->cond(), stmt->condition_position());
VisitForControl(stmt->cond(),
&stack_check,
loop_statement.break_label(),
&stack_check);
// Check stack before looping.
PrepareForBailoutForId(stmt->BackEdgeId(), NO_REGISTERS);
__ bind(&stack_check);
EmitStackCheck(stmt, &body);
__ jmp(&body);
PrepareForBailoutForId(stmt->ExitId(), NO_REGISTERS);
__ bind(loop_statement.break_label());
decrement_loop_depth();
Initial infrastructure for fast compilation of top-level code. The fast code generator is optimized for compilation time and code size. Currently it is only implemented on IA32. It is potentially triggered for any code in the global scope (including code eval'd in the global scope). It performs a syntactic check and chooses to compile in fast mode if the AST contains only supported constructs and matches some other constraints. Initially supported constructs are * ExpressionStatement, * ReturnStatement, * VariableProxy (variable references) to parameters and stack-allocated locals, * Assignment with lhs a parameter or stack-allocated local, and * Literal This allows compilation of literals at the top level and not much else. All intermediate values are allocated to temporaries and the stack is used for all temporaries. The extra memory traffic is a known issue. The code generated for 'true' is: 0 push ebp 1 mov ebp,esp 3 push esi 4 push edi 5 push 0xf5cca135 ;; object: 0xf5cca135 <undefined> 10 cmp esp,[0x8277efc] 16 jnc 27 (0xf5cbbb1b) 22 call 0xf5cac960 ;; code: STUB, StackCheck, minor: 0 27 push 0xf5cca161 ;; object: 0xf5cca161 <true> 32 mov eax,[esp] 35 mov [ebp+0xf4],eax 38 pop eax 39 mov eax,[ebp+0xf4] 42 mov esp,ebp ;; js return 44 pop ebp 45 ret 0x4 48 mov eax,0xf5cca135 ;; object: 0xf5cca135 <undefined> 53 mov esp,ebp ;; js return 55 pop ebp 56 ret 0x4 Review URL: http://codereview.chromium.org/273050 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@3067 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-10-14 19:30:50 +00:00
}
void FullCodeGenerator::VisitWhileStatement(WhileStatement* stmt) {
Comment cmnt(masm_, "[ WhileStatement");
Label test, body;
Iteration loop_statement(this, stmt);
increment_loop_depth();
// Emit the test at the bottom of the loop.
__ jmp(&test);
PrepareForBailoutForId(stmt->BodyId(), NO_REGISTERS);
__ bind(&body);
Visit(stmt->body());
// Emit the statement position here as this is where the while
// statement code starts.
__ bind(loop_statement.continue_label());
SetStatementPosition(stmt);
// Check stack before looping.
EmitStackCheck(stmt, &body);
__ bind(&test);
VisitForControl(stmt->cond(),
&body,
loop_statement.break_label(),
loop_statement.break_label());
PrepareForBailoutForId(stmt->ExitId(), NO_REGISTERS);
__ bind(loop_statement.break_label());
decrement_loop_depth();
Initial infrastructure for fast compilation of top-level code. The fast code generator is optimized for compilation time and code size. Currently it is only implemented on IA32. It is potentially triggered for any code in the global scope (including code eval'd in the global scope). It performs a syntactic check and chooses to compile in fast mode if the AST contains only supported constructs and matches some other constraints. Initially supported constructs are * ExpressionStatement, * ReturnStatement, * VariableProxy (variable references) to parameters and stack-allocated locals, * Assignment with lhs a parameter or stack-allocated local, and * Literal This allows compilation of literals at the top level and not much else. All intermediate values are allocated to temporaries and the stack is used for all temporaries. The extra memory traffic is a known issue. The code generated for 'true' is: 0 push ebp 1 mov ebp,esp 3 push esi 4 push edi 5 push 0xf5cca135 ;; object: 0xf5cca135 <undefined> 10 cmp esp,[0x8277efc] 16 jnc 27 (0xf5cbbb1b) 22 call 0xf5cac960 ;; code: STUB, StackCheck, minor: 0 27 push 0xf5cca161 ;; object: 0xf5cca161 <true> 32 mov eax,[esp] 35 mov [ebp+0xf4],eax 38 pop eax 39 mov eax,[ebp+0xf4] 42 mov esp,ebp ;; js return 44 pop ebp 45 ret 0x4 48 mov eax,0xf5cca135 ;; object: 0xf5cca135 <undefined> 53 mov esp,ebp ;; js return 55 pop ebp 56 ret 0x4 Review URL: http://codereview.chromium.org/273050 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@3067 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-10-14 19:30:50 +00:00
}
void FullCodeGenerator::VisitForStatement(ForStatement* stmt) {
Comment cmnt(masm_, "[ ForStatement");
Label test, body;
Iteration loop_statement(this, stmt);
// Set statement position for a break slot before entering the for-body.
SetStatementPosition(stmt);
if (stmt->init() != NULL) {
Visit(stmt->init());
}
increment_loop_depth();
// Emit the test at the bottom of the loop (even if empty).
__ jmp(&test);
PrepareForBailoutForId(stmt->BodyId(), NO_REGISTERS);
__ bind(&body);
Visit(stmt->body());
PrepareForBailoutForId(stmt->ContinueId(), NO_REGISTERS);
__ bind(loop_statement.continue_label());
if (stmt->next() != NULL) {
Visit(stmt->next());
}
// Emit the statement position here as this is where the for
// statement code starts.
SetStatementPosition(stmt);
// Check stack before looping.
EmitStackCheck(stmt, &body);
__ bind(&test);
if (stmt->cond() != NULL) {
VisitForControl(stmt->cond(),
&body,
loop_statement.break_label(),
loop_statement.break_label());
} else {
__ jmp(&body);
}
PrepareForBailoutForId(stmt->ExitId(), NO_REGISTERS);
__ bind(loop_statement.break_label());
decrement_loop_depth();
Initial infrastructure for fast compilation of top-level code. The fast code generator is optimized for compilation time and code size. Currently it is only implemented on IA32. It is potentially triggered for any code in the global scope (including code eval'd in the global scope). It performs a syntactic check and chooses to compile in fast mode if the AST contains only supported constructs and matches some other constraints. Initially supported constructs are * ExpressionStatement, * ReturnStatement, * VariableProxy (variable references) to parameters and stack-allocated locals, * Assignment with lhs a parameter or stack-allocated local, and * Literal This allows compilation of literals at the top level and not much else. All intermediate values are allocated to temporaries and the stack is used for all temporaries. The extra memory traffic is a known issue. The code generated for 'true' is: 0 push ebp 1 mov ebp,esp 3 push esi 4 push edi 5 push 0xf5cca135 ;; object: 0xf5cca135 <undefined> 10 cmp esp,[0x8277efc] 16 jnc 27 (0xf5cbbb1b) 22 call 0xf5cac960 ;; code: STUB, StackCheck, minor: 0 27 push 0xf5cca161 ;; object: 0xf5cca161 <true> 32 mov eax,[esp] 35 mov [ebp+0xf4],eax 38 pop eax 39 mov eax,[ebp+0xf4] 42 mov esp,ebp ;; js return 44 pop ebp 45 ret 0x4 48 mov eax,0xf5cca135 ;; object: 0xf5cca135 <undefined> 53 mov esp,ebp ;; js return 55 pop ebp 56 ret 0x4 Review URL: http://codereview.chromium.org/273050 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@3067 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-10-14 19:30:50 +00:00
}
void FullCodeGenerator::VisitTryCatchStatement(TryCatchStatement* stmt) {
Comment cmnt(masm_, "[ TryCatchStatement");
SetStatementPosition(stmt);
// The try block adds a handler to the exception handler chain before
// entering, and removes it again when exiting normally. If an exception
// is thrown during execution of the try block, the handler is consumed
// and control is passed to the catch block with the exception in the
// result register.
Label try_entry, handler_entry, exit;
__ jmp(&try_entry);
__ bind(&handler_entry);
handler_table()->set(stmt->index(), Smi::FromInt(handler_entry.pos()));
// Exception handler code, the exception is in the result register.
// Extend the context before executing the catch block.
{ Comment cmnt(masm_, "[ Extend catch context");
__ Push(stmt->variable()->name());
__ push(result_register());
PushFunctionArgumentForContextAllocation();
__ CallRuntime(Runtime::kPushCatchContext, 3);
StoreToFrameField(StandardFrameConstants::kContextOffset,
context_register());
}
Scope* saved_scope = scope();
scope_ = stmt->scope();
ASSERT(scope_->declarations()->is_empty());
{ WithOrCatch catch_body(this);
Visit(stmt->catch_block());
}
// Restore the context.
LoadContextField(context_register(), Context::PREVIOUS_INDEX);
StoreToFrameField(StandardFrameConstants::kContextOffset, context_register());
scope_ = saved_scope;
__ jmp(&exit);
// Try block code. Sets up the exception handler chain.
__ bind(&try_entry);
__ PushTryHandler(StackHandler::CATCH, stmt->index());
{ TryCatch try_body(this);
Visit(stmt->try_block());
}
__ PopTryHandler();
__ bind(&exit);
Initial infrastructure for fast compilation of top-level code. The fast code generator is optimized for compilation time and code size. Currently it is only implemented on IA32. It is potentially triggered for any code in the global scope (including code eval'd in the global scope). It performs a syntactic check and chooses to compile in fast mode if the AST contains only supported constructs and matches some other constraints. Initially supported constructs are * ExpressionStatement, * ReturnStatement, * VariableProxy (variable references) to parameters and stack-allocated locals, * Assignment with lhs a parameter or stack-allocated local, and * Literal This allows compilation of literals at the top level and not much else. All intermediate values are allocated to temporaries and the stack is used for all temporaries. The extra memory traffic is a known issue. The code generated for 'true' is: 0 push ebp 1 mov ebp,esp 3 push esi 4 push edi 5 push 0xf5cca135 ;; object: 0xf5cca135 <undefined> 10 cmp esp,[0x8277efc] 16 jnc 27 (0xf5cbbb1b) 22 call 0xf5cac960 ;; code: STUB, StackCheck, minor: 0 27 push 0xf5cca161 ;; object: 0xf5cca161 <true> 32 mov eax,[esp] 35 mov [ebp+0xf4],eax 38 pop eax 39 mov eax,[ebp+0xf4] 42 mov esp,ebp ;; js return 44 pop ebp 45 ret 0x4 48 mov eax,0xf5cca135 ;; object: 0xf5cca135 <undefined> 53 mov esp,ebp ;; js return 55 pop ebp 56 ret 0x4 Review URL: http://codereview.chromium.org/273050 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@3067 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-10-14 19:30:50 +00:00
}
void FullCodeGenerator::VisitTryFinallyStatement(TryFinallyStatement* stmt) {
Comment cmnt(masm_, "[ TryFinallyStatement");
SetStatementPosition(stmt);
// Try finally is compiled by setting up a try-handler on the stack while
// executing the try body, and removing it again afterwards.
//
// The try-finally construct can enter the finally block in three ways:
// 1. By exiting the try-block normally. This removes the try-handler and
// calls the finally block code before continuing.
// 2. By exiting the try-block with a function-local control flow transfer
// (break/continue/return). The site of the, e.g., break removes the
// try handler and calls the finally block code before continuing
// its outward control transfer.
// 3. By exiting the try-block with a thrown exception.
// This can happen in nested function calls. It traverses the try-handler
// chain and consumes the try-handler entry before jumping to the
// handler code. The handler code then calls the finally-block before
// rethrowing the exception.
//
// The finally block must assume a return address on top of the stack
// (or in the link register on ARM chips) and a value (return value or
// exception) in the result register (rax/eax/r0), both of which must
// be preserved. The return address isn't GC-safe, so it should be
// cooked before GC.
Label try_entry, handler_entry, finally_entry;
// Jump to try-handler setup and try-block code.
__ jmp(&try_entry);
__ bind(&handler_entry);
handler_table()->set(stmt->index(), Smi::FromInt(handler_entry.pos()));
// Exception handler code. This code is only executed when an exception
// is thrown. The exception is in the result register, and must be
// preserved by the finally block. Call the finally block and then
// rethrow the exception if it returns.
__ Call(&finally_entry);
__ push(result_register());
__ CallRuntime(Runtime::kReThrow, 1);
// Finally block implementation.
__ bind(&finally_entry);
EnterFinallyBlock();
{ Finally finally_body(this);
Visit(stmt->finally_block());
}
ExitFinallyBlock(); // Return to the calling code.
// Set up try handler.
__ bind(&try_entry);
__ PushTryHandler(StackHandler::FINALLY, stmt->index());
{ TryFinally try_body(this, &finally_entry);
Visit(stmt->try_block());
}
__ PopTryHandler();
// Execute the finally block on the way out. Clobber the unpredictable
// value in the result register with one that's safe for GC because the
// finally block will unconditionally preserve the result register on the
// stack.
ClearAccumulator();
__ Call(&finally_entry);
Initial infrastructure for fast compilation of top-level code. The fast code generator is optimized for compilation time and code size. Currently it is only implemented on IA32. It is potentially triggered for any code in the global scope (including code eval'd in the global scope). It performs a syntactic check and chooses to compile in fast mode if the AST contains only supported constructs and matches some other constraints. Initially supported constructs are * ExpressionStatement, * ReturnStatement, * VariableProxy (variable references) to parameters and stack-allocated locals, * Assignment with lhs a parameter or stack-allocated local, and * Literal This allows compilation of literals at the top level and not much else. All intermediate values are allocated to temporaries and the stack is used for all temporaries. The extra memory traffic is a known issue. The code generated for 'true' is: 0 push ebp 1 mov ebp,esp 3 push esi 4 push edi 5 push 0xf5cca135 ;; object: 0xf5cca135 <undefined> 10 cmp esp,[0x8277efc] 16 jnc 27 (0xf5cbbb1b) 22 call 0xf5cac960 ;; code: STUB, StackCheck, minor: 0 27 push 0xf5cca161 ;; object: 0xf5cca161 <true> 32 mov eax,[esp] 35 mov [ebp+0xf4],eax 38 pop eax 39 mov eax,[ebp+0xf4] 42 mov esp,ebp ;; js return 44 pop ebp 45 ret 0x4 48 mov eax,0xf5cca135 ;; object: 0xf5cca135 <undefined> 53 mov esp,ebp ;; js return 55 pop ebp 56 ret 0x4 Review URL: http://codereview.chromium.org/273050 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@3067 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-10-14 19:30:50 +00:00
}
void FullCodeGenerator::VisitDebuggerStatement(DebuggerStatement* stmt) {
#ifdef ENABLE_DEBUGGER_SUPPORT
Comment cmnt(masm_, "[ DebuggerStatement");
SetStatementPosition(stmt);
__ DebugBreak();
// Ignore the return value.
#endif
Initial infrastructure for fast compilation of top-level code. The fast code generator is optimized for compilation time and code size. Currently it is only implemented on IA32. It is potentially triggered for any code in the global scope (including code eval'd in the global scope). It performs a syntactic check and chooses to compile in fast mode if the AST contains only supported constructs and matches some other constraints. Initially supported constructs are * ExpressionStatement, * ReturnStatement, * VariableProxy (variable references) to parameters and stack-allocated locals, * Assignment with lhs a parameter or stack-allocated local, and * Literal This allows compilation of literals at the top level and not much else. All intermediate values are allocated to temporaries and the stack is used for all temporaries. The extra memory traffic is a known issue. The code generated for 'true' is: 0 push ebp 1 mov ebp,esp 3 push esi 4 push edi 5 push 0xf5cca135 ;; object: 0xf5cca135 <undefined> 10 cmp esp,[0x8277efc] 16 jnc 27 (0xf5cbbb1b) 22 call 0xf5cac960 ;; code: STUB, StackCheck, minor: 0 27 push 0xf5cca161 ;; object: 0xf5cca161 <true> 32 mov eax,[esp] 35 mov [ebp+0xf4],eax 38 pop eax 39 mov eax,[ebp+0xf4] 42 mov esp,ebp ;; js return 44 pop ebp 45 ret 0x4 48 mov eax,0xf5cca135 ;; object: 0xf5cca135 <undefined> 53 mov esp,ebp ;; js return 55 pop ebp 56 ret 0x4 Review URL: http://codereview.chromium.org/273050 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@3067 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-10-14 19:30:50 +00:00
}
void FullCodeGenerator::VisitConditional(Conditional* expr) {
Comment cmnt(masm_, "[ Conditional");
Label true_case, false_case, done;
VisitForControl(expr->condition(), &true_case, &false_case, &true_case);
PrepareForBailoutForId(expr->ThenId(), NO_REGISTERS);
__ bind(&true_case);
SetExpressionPosition(expr->then_expression(),
expr->then_expression_position());
if (context()->IsTest()) {
const TestContext* for_test = TestContext::cast(context());
VisitForControl(expr->then_expression(),
for_test->true_label(),
for_test->false_label(),
NULL);
} else {
VisitInDuplicateContext(expr->then_expression());
__ jmp(&done);
}
PrepareForBailoutForId(expr->ElseId(), NO_REGISTERS);
__ bind(&false_case);
SetExpressionPosition(expr->else_expression(),
expr->else_expression_position());
VisitInDuplicateContext(expr->else_expression());
// If control flow falls through Visit, merge it with true case here.
if (!context()->IsTest()) {
__ bind(&done);
}
Initial infrastructure for fast compilation of top-level code. The fast code generator is optimized for compilation time and code size. Currently it is only implemented on IA32. It is potentially triggered for any code in the global scope (including code eval'd in the global scope). It performs a syntactic check and chooses to compile in fast mode if the AST contains only supported constructs and matches some other constraints. Initially supported constructs are * ExpressionStatement, * ReturnStatement, * VariableProxy (variable references) to parameters and stack-allocated locals, * Assignment with lhs a parameter or stack-allocated local, and * Literal This allows compilation of literals at the top level and not much else. All intermediate values are allocated to temporaries and the stack is used for all temporaries. The extra memory traffic is a known issue. The code generated for 'true' is: 0 push ebp 1 mov ebp,esp 3 push esi 4 push edi 5 push 0xf5cca135 ;; object: 0xf5cca135 <undefined> 10 cmp esp,[0x8277efc] 16 jnc 27 (0xf5cbbb1b) 22 call 0xf5cac960 ;; code: STUB, StackCheck, minor: 0 27 push 0xf5cca161 ;; object: 0xf5cca161 <true> 32 mov eax,[esp] 35 mov [ebp+0xf4],eax 38 pop eax 39 mov eax,[ebp+0xf4] 42 mov esp,ebp ;; js return 44 pop ebp 45 ret 0x4 48 mov eax,0xf5cca135 ;; object: 0xf5cca135 <undefined> 53 mov esp,ebp ;; js return 55 pop ebp 56 ret 0x4 Review URL: http://codereview.chromium.org/273050 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@3067 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-10-14 19:30:50 +00:00
}
void FullCodeGenerator::VisitLiteral(Literal* expr) {
Comment cmnt(masm_, "[ Literal");
context()->Plug(expr->handle());
}
void FullCodeGenerator::VisitFunctionLiteral(FunctionLiteral* expr) {
Comment cmnt(masm_, "[ FunctionLiteral");
// Build the function boilerplate and instantiate it.
Handle<SharedFunctionInfo> function_info =
Compiler::BuildFunctionInfo(expr, script());
if (function_info.is_null()) {
SetStackOverflow();
return;
}
EmitNewClosure(function_info, expr->pretenure());
}
void FullCodeGenerator::VisitSharedFunctionInfoLiteral(
SharedFunctionInfoLiteral* expr) {
Comment cmnt(masm_, "[ SharedFunctionInfoLiteral");
EmitNewClosure(expr->shared_function_info(), false);
}
void FullCodeGenerator::VisitThrow(Throw* expr) {
Comment cmnt(masm_, "[ Throw");
VisitForStackValue(expr->exception());
__ CallRuntime(Runtime::kThrow, 1);
// Never returns here.
Initial infrastructure for fast compilation of top-level code. The fast code generator is optimized for compilation time and code size. Currently it is only implemented on IA32. It is potentially triggered for any code in the global scope (including code eval'd in the global scope). It performs a syntactic check and chooses to compile in fast mode if the AST contains only supported constructs and matches some other constraints. Initially supported constructs are * ExpressionStatement, * ReturnStatement, * VariableProxy (variable references) to parameters and stack-allocated locals, * Assignment with lhs a parameter or stack-allocated local, and * Literal This allows compilation of literals at the top level and not much else. All intermediate values are allocated to temporaries and the stack is used for all temporaries. The extra memory traffic is a known issue. The code generated for 'true' is: 0 push ebp 1 mov ebp,esp 3 push esi 4 push edi 5 push 0xf5cca135 ;; object: 0xf5cca135 <undefined> 10 cmp esp,[0x8277efc] 16 jnc 27 (0xf5cbbb1b) 22 call 0xf5cac960 ;; code: STUB, StackCheck, minor: 0 27 push 0xf5cca161 ;; object: 0xf5cca161 <true> 32 mov eax,[esp] 35 mov [ebp+0xf4],eax 38 pop eax 39 mov eax,[ebp+0xf4] 42 mov esp,ebp ;; js return 44 pop ebp 45 ret 0x4 48 mov eax,0xf5cca135 ;; object: 0xf5cca135 <undefined> 53 mov esp,ebp ;; js return 55 pop ebp 56 ret 0x4 Review URL: http://codereview.chromium.org/273050 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@3067 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-10-14 19:30:50 +00:00
}
FullCodeGenerator::NestedStatement* FullCodeGenerator::TryCatch::Exit(
int* stack_depth,
int* context_length) {
// The macros used here must preserve the result register.
__ Drop(*stack_depth);
__ PopTryHandler();
*stack_depth = 0;
return previous_;
}
2011-09-19 14:50:33 +00:00
bool FullCodeGenerator::TryLiteralCompare(CompareOperation* expr) {
Expression* sub_expr;
Handle<String> check;
2011-09-19 14:50:33 +00:00
if (expr->IsLiteralCompareTypeof(&sub_expr, &check)) {
EmitLiteralCompareTypeof(expr, sub_expr, check);
return true;
}
2011-09-19 14:50:33 +00:00
if (expr->IsLiteralCompareUndefined(&sub_expr)) {
EmitLiteralCompareNil(expr, sub_expr, kUndefinedValue);
return true;
}
2011-09-19 14:50:33 +00:00
if (expr->IsLiteralCompareNull(&sub_expr)) {
EmitLiteralCompareNil(expr, sub_expr, kNullValue);
return true;
}
return false;
}
#undef __
Initial infrastructure for fast compilation of top-level code. The fast code generator is optimized for compilation time and code size. Currently it is only implemented on IA32. It is potentially triggered for any code in the global scope (including code eval'd in the global scope). It performs a syntactic check and chooses to compile in fast mode if the AST contains only supported constructs and matches some other constraints. Initially supported constructs are * ExpressionStatement, * ReturnStatement, * VariableProxy (variable references) to parameters and stack-allocated locals, * Assignment with lhs a parameter or stack-allocated local, and * Literal This allows compilation of literals at the top level and not much else. All intermediate values are allocated to temporaries and the stack is used for all temporaries. The extra memory traffic is a known issue. The code generated for 'true' is: 0 push ebp 1 mov ebp,esp 3 push esi 4 push edi 5 push 0xf5cca135 ;; object: 0xf5cca135 <undefined> 10 cmp esp,[0x8277efc] 16 jnc 27 (0xf5cbbb1b) 22 call 0xf5cac960 ;; code: STUB, StackCheck, minor: 0 27 push 0xf5cca161 ;; object: 0xf5cca161 <true> 32 mov eax,[esp] 35 mov [ebp+0xf4],eax 38 pop eax 39 mov eax,[ebp+0xf4] 42 mov esp,ebp ;; js return 44 pop ebp 45 ret 0x4 48 mov eax,0xf5cca135 ;; object: 0xf5cca135 <undefined> 53 mov esp,ebp ;; js return 55 pop ebp 56 ret 0x4 Review URL: http://codereview.chromium.org/273050 git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@3067 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-10-14 19:30:50 +00:00
} } // namespace v8::internal